Enhance merge_rotations transform

[andrijapau]

Feature details

We currently have the behaviour,

@qml.transforms.optimization.merge_rotations
@qml.qnode(dev)
def qfunc_decomposed():
    qml.RZ(-1., 0)
    qml.RY(0, 0)
    qml.RZ(1, 0)
    qml.RZ(-1, 0)
    qml.RY(0, 0)
    qml.RZ(1., 0)

>>> print(qml.draw(qfunc_decomposed)())
0: ──RZ(-1.00)──RZ(1.00)─┤ # Should cancel! 

Which suggests the algorithm is unable to collapse and merge rotation gates correctly. This can be remedied by another pass with the transform but ideally it should happen all at once.

Implementation

Requires modifying the algorithmic implementation found in pennylane/transforms/optimization/merge_rotations.py.

How important would you say this feature is?

1: Not important. Would be nice to have.

Additional information

Once the enhancement has been made, relevant tests should be added.

[andrijapau]

Closing this issue after internal discussion. 🥳

Summary: According to the docstring of this transform,

"""Quantum transform to combine rotation gates of the same type that act sequentially."""

this is expected behaviour. If the user wishes to merge all of them they would have to use single_qubit_fusion or multiple compilation passes with this transform. 🙂