alignment: add plugin test suite with Halton sky sampling

[ckemper67]

Summary

This branch establishes a test harness for INDI alignment math plugins and fixes two correctness bugs in the simulator and BuiltIn plugin that the tests exposed.

Interface changes

• MathPlugin::TransformTelescopeToCelestial gains a JulianOffset = 0 parameter, matching the existing TransformCelestialToTelescope signature. This allows tests to inject a fixed Julian Date and get deterministic transforms independent of the system clock. All existing implementations updated; a TODO notes that a future refactor should move to an absolute Julian Date rather than an offset.
• InMemoryDatabase::SetDatabaseReferencePosition gains an IGeographicCoordinates overload alongside the existing (double lat, double lon) form.
• indicom.h gains angle-unit conversion macros: HOURS_TO_DEG, DEG_TO_HOURS, ARCMIN_TO_DEG, DEG_TO_ARCMIN, ARCSEC_TO_DEG, DEG_TO_ARCSEC, HOURS_TO_RAD, RAD_TO_HOURS (all guarded with #ifndef).
• BuiltInMathPlugin gains a NO_PLUGIN_HOOKS guard around its extern "C" factory functions, matching the pattern used by SVD and Nearest, so the translation unit can be linked directly into test binaries without symbol conflicts.

Bug fixes

scopesim_helper.cpp — AltAz azimuth convention
apparentHaDecToMount was producing azimuth in scopesim's internal convention (0° = South, clockwise) without converting to INDI's IHorizontalCoordinates convention (0° = North). Added +Angle(180.0) to the primary output and a return to prevent fallthrough into the equatorial code path.

scopesim_helper.cpp — GEM meridian flip
mountToApparentHaDec initialised prio inside an else branch, leaving it uninitialised when the flip condition was true. Moved prio = primary before the flip check so both branches are always covered.

libs/alignment/BuiltInMathPlugin.cpp — singular matrix fallthrough
CalculateTransformMatrices wrapped the MatrixMatrixMultiply and inverse steps in an else guarded by the singularity check, so a singular Alpha matrix silently skipped the Beta matrix calculation entirely. Removed the else; the identity fallback is set and computation continues regardless.

Test suite (test/alignment/test_alignment_plugins.cpp)

34 tests using a dual-Halton sequence design: alignment points drawn from Halton(2,3), independent validation points from Halton(5,7). This eliminates the correlation between training and test sets while giving good sky coverage with few points.

Coverage includes:

• SVDMathPlugin: polar errors, axis errors (IH/ID/CH), non-perpendicularity, small/large errors, all-terms combined, multiple geographic sites (LA, Tokyo, Sydney, Nairobi), AltAz, EQ_FORK, GEM meridian flip (north and south), convex hull fallback
• BuiltInMathPlugin: all-terms equatorial, AltAz, meridian flip
• NearestMathPlugin: round-trip (all-in-alignment), coarse quality check, AltAz

Assertions use a fixed targetRMSArcsec quality target rather than a fraction of injected error, with a P95 bound at 1.5× the RMS target to catch outliers.

Test plan

cd build/test/alignment
cmake --build . --target test_alignment_plugins
./test_alignment_plugins

Expected: 34 tests pass. The two DISABLED_Nearest_MeridianFlip tests are intentionally skipped (NearestMathPlugin does not handle flip geometry reliably without dense coverage).

[knro]

Thank you for the changes. Are you using the Alignment module for any specific requirement when you ran into these issues?

[ckemper67]

Thank you for the changes. Are you using the Alignment module for any specific requirement when you ran into these issues?

Ah sorry. The context is in a message on the indi dev forum. I am writing a new math plugin and when I wrote the tests for it they uncovered some issues in the existing plugins.