Fix handling of cumulative_angles in merge_rotations

[andrijapau]

Context:

For the Rot gate, the angles can cancel in a non-trivial way (e.g. Rot(φ,0,-φ) = RZ(φ) RY(0) RZ(-φ) = RZ(0) = I).

Unlike single_qubit_fusion, this was not being handled correctly and only the trivial case where all angles are zero was being checked.

Description of the Change:

• Create a boolean flag angles_cancel that determines whether the operator is the identity and can be ignored in the transformed tape.
• Added a test to cover the use case.

Benefits: Correct behaviour.

Possible Drawbacks: None identified.

Related GitHub Issues: Fixes #7010

[sc-85327]

[andrijapau]

@albi3ro, @dwierichs

I've deleted the xfaild test associated with this issue #7009 after the discussion that this is expected behaviour for the merge_rotations transform. 👍🏼

[dwierichs]

Just had a minor suggestion for readability, otherwise LGTM 🚀
Thanks for this upgrade @andrijapau 🎉

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.60%. Comparing base (a0f6f71) to head (743f737).
Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #7011   +/-   ##
=======================================
  Coverage   99.60%   99.60%           
=======================================
  Files         485      485           
  Lines       46361    46365    +4     
=======================================
+ Hits        46178    46182    +4     
  Misses        183      183           

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