Add requirements.txt, missing patch to ASV benchmarks

[vchamarthi]

• Add _patch_setup.py that calls patch_numpy_fft() at worker import time with hard-fail validation (ensures benchmarks never silently run with stock NumPy)
• Add requirements.txt with psutil and scipy benchmark dependencies
• Update the regression threshold matching the mkl_umath

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR updates the benchmarks setup to force NumPy’s FFT to run through Intel MKL (via mkl_fft) and tightens ASV regression thresholds.

Changes:

• Add an import-time patch step that applies mkl_fft.patch_numpy_fft() during benchmark initialization.
• Add a benchmarks requirements.txt with baseline dependencies.
• Reduce ASV regressions_thresholds from 0.3 to 0.2.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
benchmarks/requirements.txt	Adds benchmarks dependencies list for environment creation.
benchmarks/benchmarks/_patch_setup.py	Implements mkl_fft patching with hard-fail checks and diagnostics.
benchmarks/benchmarks/init.py	Calls the patch function during benchmark package import.
benchmarks/asv.conf.json	Tightens regression threshold to catch smaller performance changes.

[ndgrigorian]

making this a dictionary would make more sense to me, especially with the name "PATCH_MAP"

[ndgrigorian]

Suggested change

	f"[mkl-patch] {mod_name} has no {patch_fn_name}(). "
	f"{mod_name} has no {patch_fn_name}(). "

[ndgrigorian]

all [mkl-patch] should be removed I feel, it's a bit confusing, makes it seem like a package (we have mkl-service after all).

[ndgrigorian]

I think we can drop this line -- the patching already has a verbose mode. If anything, we could just call into the verbose mode, I feel.

[ndgrigorian]

same here, a call with verbose patching could substitute