Use the current interpreter in the comparative benchmark runner

[ayaangazali]

Proposed changes

benchmarks/python/comparative/compare.py shells out to a literal "python", so it fails immediately on any setup where only python3 is on PATH, which includes a stock macOS shell:

$ python benchmarks/python/comparative/compare.py ...
FileNotFoundError: [Errno 2] No such file or directory: 'python'

Switched the four subprocess calls to sys.executable. Besides fixing the crash, this makes the child benchmarks run under the same interpreter as the parent, so running compare.py from a virtualenv now benchmarks that environment's MLX rather than whatever python happens to resolve to.

mlx.launch already does the same thing for the same reason (--python defaults to sys.executable).

After the change the subprocesses actually start, which you can see because bench_mlx.py gets far enough to parse its own arguments. On this machine the run then stops in bench_torch.py because torch is not installed, which is the expected behaviour for a comparison benchmark without torch.

Checklist

Put an x in the boxes that apply.

• I have read the CONTRIBUTING document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)

(no tests since this is a benchmark helper script; verified by running it before and after)

---

as always, being clear about process: freshman contributor and i use Claude Code to help me look around, but i hit this by actually running the benchmark scripts, and i checked that sys.executable is the pattern already used elsewhere in the repo before picking it. i also grepped for other hardcoded "python" subprocess calls and did not find any, so this looks like the only spot.