Fix runtime_artifacts walk performance

[mehrdad2m]

Context:
Currently there is a performance degradation issue with catalyst.runtime_artifacts where it walks the entire IR operations looking for runtime_artifacts which is unnecessary.

Description of the Change:
It now visits module operations only, instead of every operation in the program.

Benefits:

Possible Drawbacks:

Related GitHub Issues:
[sc-129402]

[mehrdad2m]

!benchmark

[dime10]

What I don't understand is how a single walk could produce a meaningful degradation, since we do do walks quite often (although I agree recursive walks should be avoided if unnecessary) 🤔

[mehrdad2m]

!benchmark

[jzaia18]

!benchmark

[ringo-but-quantum]

Hi @jzaia18, thanks for your benchmark request! We have received the request to benchmark this PR. We will start running the benchmarks shortly and will add a new comment with the results once they are ready.

See this README section for more information on how the benchmarks are triggered and run.

Reports to expect:

• A report analyzing the performance of 'lightning.qubit'.
• A report analyzing the performance of 'null.qubit'.

You can track the progress of the benchmarks here.