Various minor fixes to the Matlab toolbox - #2168

The only behavior this overload needs to define is the function-call form
f(x). The '.' branch and the leading length(s) > 1 chaining block were
hand-rolled reimplementations of default subsref behavior, and worse at it
than the default: the chaining block assumed s(1).type was '.' without
checking, so f(2).type failed on a cell-array field name, and the
fall-through error ('Specify value for x as p(x)') was misleading for a
brace reference.

Handle '()' and pass everything else to builtin('subsref', obj, s), with a
comment saying why, and trim the docstring to document the interesting case
instead of restating default behavior.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The functor-type bullet list in the constructor docstring was indented
past the surrounding text, nesting it in a blockquote, and its usage
examples lacked the '::' marker, so they rendered as running text with
smart quotes instead of literal blocks.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

ct.oneD.Flow is the base class for the flow domains but was missing from
the Sphinx listing, so its properties and methods were undocumented and
the concrete flow domains showed no base class.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

auto_path.glob() returns a generator that was fully consumed by the loop
building the list of generated files, so env.Depends() was handed an
empty list and edits to interfaces/sourcegen/.../headers/ct*.yaml never
triggered regeneration of the CLib sources.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The method passed an extra trailing argument to surf_setCoverages, which
does not match the generated CLib signature and would not have reached
SurfPhase::setCoveragesNoNorm in any case. Add a CLib recipe for
setCoveragesNoNorm and call it, with tests on both sides.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

CLib provided only the name-based overload, so the index-based form was
unreachable from MATLAB and Julia. Following the sol_adjacent /
sol_adjacentByName precedent, reactornet_sensitivity now wraps
sensitivity(size_t, size_t) and the name-keyed form becomes
reactornet_sensitivityByName.

MATLAB's ReactorNet.sensitivity now dispatches on the type of the
component argument, errors on an unsupported type, and passes a reactor
index rather than a cabinet handle; it accepts a reactor object or a
1-based position, and ReactorNet gains a reactors property to support
the lookup. Julia gains the index-based method.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The toolbox provided MassFlowController and Valve but no
PressureController. Two bugs blocked it: FlowDevice.setPrimary passed the
device object where CLib expects a handle, and the only coefficient
setter refused anything that was not a Valve. Add a generic
deviceCoefficient property that valveCoeff and the new pressureCoeff
both delegate to.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The method was declared with an output argument that its body never
assigned, so any caller asking for one got "Output argument 'tp' not
assigned" rather than the copy the signature implied. equilibrate
modifies the phase in place; declare it with no output.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The method returned the CLib status code and documented it as "The error
in the solution". It is neither: MultiPhase::equilibrate returns void,
mix_equilibrate returns 0 on success and -1 on failure, and ct.impl.call
already turns the failure code into a Cantera:ctError. The value handed
back was therefore an unconditional zero described as a solver residual.

The claim predates the toolbox rewrite -- the legacy MEX interface
removed in 5e275fc carried the same docstring over the same
always-zero return. Declare the method with no output, matching
ThermoPhase.equilibrate and Python's Mixture.equilibrate.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

get.massFlowRate declared its output argument as "flag" but assigned the
CLib result to a local "rate", so the output was never set and every read
of the property raised MATLAB:unassignedOutputs.

set.massFlowRate also assigned obj.massFlowRate at the end. That is not
recursive -- MATLAB suppresses a property's set method for an assignment
made inside that same method -- but it is dead: get.massFlowRate shadows
the stored value, so nothing could ever read it. Drop it, matching the
forward-only set methods in ReactorBase.

Expand the property documentation to describe both directions, following
FlowReactor::massFlowRate (m_u * m_rho * m_area) and setMassFlowRate,
which sets the speed from the current density and area.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The docstring for the coverageEnabled property was copied from an
unrelated method: it described "setting bounds on the solution
components" and documented a `:param flag:` for a call form rather than
describing the property itself. Replace it with a description of what
the flag actually controls, matching the Python `coverage_enabled`
docstring.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The default steady-state tolerances and maximum Jacobian age are good enough
for these examples, so the explicit setSteadyTolerances, setTransientTolerances
and setMaxJacAge calls were just noise for readers. The equivalent Python
examples set none of them.

diffusion_flame is the exception: it still needs a relaxed transient absolute
tolerance, since the default is too tight for the time stepping used to get the
flame started and the solver stalls at its minimum timestep. Only that one call
is kept, with the relative tolerance left at its default and a comment saying why.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The `concentrations` property was reported broken in the experimental
toolbox. It works on main: the getter read the object handle from an
undefined `tp` between 4d0aa3b and f5811df, a window that only the
v3.2.0b1 tag ever shipped. Cover both the bulk and the surface phase so
the regression cannot come back unnoticed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The docstring promised a cell array of directories, but the function
returned the single PATH-like string that CLib builds, and it hard-coded
';' as the separator handed to Cantera::getDataDirectories -- wrong on
every platform where pathsep is ':'.

Pass pathsep in both directions and split the result, so the join and the
split cannot drift apart. This matches get_data_directories() in the
Python module, which uses os.pathsep the same way.

Add ctTestUtility, a home for the package-level ct.* utility functions,
covering the shape of the returned cell array.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Neither ct.zeroD.ReactorBase nor ct.zeroD.ReactorNet offered any way to map
between a state-vector index and a component name, so MATLAB code needing a
global index had to hard-code the state-vector layout of the reactor type it
was using.

Add four CLib recipes -- reactor_neq, reactor_componentName,
reactor_componentIndex and reactornet_globalComponentIndex -- and the six
corresponding MATLAB members: ReactorBase.nVars, ReactorBase.componentIndex,
ReactorBase.componentName, ReactorNet.nVars, ReactorNet.componentName and
ReactorNet.globalComponentIndex. All MATLAB indices are 1-based in both
directions. globalComponentIndex takes its reactor argument in the same forms
as ReactorNet.sensitivity, and that dispatch is now a shared private helper
that also range-checks a numeric index; without it, an out-of-range index
reaches an unchecked m_reactors[reactor] in C++.

Note that ReactorNet.componentName returns a reactor-prefixed name such as
'reactor1: temperature', while globalComponentIndex takes the unprefixed name
resolved within a given reactor; the two are not inverses.

The sensitivity tests added alongside the ReactorNet.sensitivity work now take
their global indices from globalComponentIndex rather than from the literals
3 and 3 + nSpecies.

No C++ change is required.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

`clib_defs.h` defines four values that a generated CLib function can return to
signal that an exception was raised: -1 and -2 (for functions whose normal
return value is a length or a handle), `DERR` (-999.999), and `ERR` (-999).
`ct.impl.errorCode` listed every one of these but `ERR`, so a function
generated as a "size getter" -- `domain_componentIndex`, `reactor_neq`,
`reactornet_globalComponentIndex` and the like -- returned -999 to MATLAB as if
it were a valid result, and the exception was silently swallowed.

Add -999 to the list, and drop the two local workarounds in
`ct.zeroD.ReactorBase.componentIndex` and
`ct.zeroD.ReactorNet.globalComponentIndex` that this makes dead. The Cantera
error message is unchanged; it now comes from `ct.impl.call` instead.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>