[Decomposition] Custom decomposition rules for symbolic operators

[astralcai]

Context:

Description of the Change:

Symbolic operator types can now be given as strings to the op_type argument of add_decomps or as keys of the dictionaries passed to the alt_decomps and fixed_decomps arguments of the decompose transform, allowing custom decomposition rules to be defined and registered for symbolic operators.

@register_resources({qml.RY: 1})
def my_adjoint_ry(phi, wires, **_):
    qml.RY(-phi, wires=wires)

@qml.register_resources({qml.RX: 1})
def my_adjoint_rx(phi, wires, **__):
    qml.RX(-phi, wires)

# Registers a decomposition rule for the adjoint of RY globally
qml.add_decomps("Adjoint(RY)", my_adjoint_ry)

@partial(
    qml.transforms.decompose,
    gate_set={"RX", "CNOT"},
    fixed_decomps={"Adjoint(RX)": my_adjoint_rx}
)
@qml.qnode(qml.device("default.qubit"))
def circuit():
    qml.adjoint(qml.RX(0.5), wires=[0])
    qml.CNOT(wires=[0, 1])
    qml.adjoint(qml.RY(0.5), wires=[1])
    return qml.expval(qml.Z(0))

>>> print(qml.draw(circuit)())
0: ──RX(-0.50)─╭●────────────┤  <Z>
1: ────────────╰X──RY(-0.50)─┤

Benefits:

Possible Drawbacks:

Related GitHub Issues:

[sc-89297]

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.65%. Comparing base (353fb36) to head (220f671).
Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #7347      +/-   ##
==========================================
- Coverage   99.65%   99.65%   -0.01%     
==========================================
  Files         529      529              
  Lines       50603    50600       -3     
==========================================
- Hits        50427    50424       -3     
  Misses        176      176              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[PietropaoloFrisoni]

I'll complete the review as soon as possible (need to interrupt for a few minutes :) )

[JerryChen97]

I don't have more questions here! Just a good-to-have non-blocking comment

[albi3ro]

We do we use issubclass for adjoint and pow but op_type in for controlled? Seems like it might be simpler to use the same pattern for all three categories.

[albi3ro]

Note that AdjointOpObs and PowOpObs will be going away in July, so we can potentially make decisions based around it being removed.

[astralcai]

For Controlled, we don't want any of the custom controlled operations like CH, CNOT, etc. to take this branch, but for Adjoint, we want all Adjoint ops including AdjointOperation to take this branch.

[albi3ro]

What about using something like resource_op when we are dealing with a CompressedResourceOp. Might make it a little clearer what of object we are dealing with, as I tend to associate op with Operator.

[astralcai]

Sure

[astralcai]

I'm renaming them back to op_node for consistency in the context of the graph.