Fix mutable default argument in autoquant.py

[petrex]

TLDR: Fix mutable default argument in autoquant.py

Fix Summary

• AutoQuantizableLinearWeight and the various _autoquant_test implementations used ["relu", None] as a default parameter list. Because lists are mutable and default parameters are evaluated once at function definition time, every call shared the exact same list object. Any in-place mutation would accidentally leak between callers.
• Added _DEFAULT_MODE_SENTINEL and _resolve_autoquant_mode() so the public API still behaves as though the default is ["relu", None], but a fresh list is created whenever a caller doesn’t supply mode.
• Constructors and _autoquant_test overrides now default to the sentinel and immediately call _resolve_autoquant_mode(mode), ensuring explicit None stays None while the implicit default gets rebuilt per call.

Expected Outcome

• Each use of autoquant mode receives its own list instance, eliminating shared state. Any code that mutates mode (e.g., mode[0] = "interpolate") affects only that invocation, preventing subtle cross-layer or cross-run contamination.
• Users should see consistent behavior across runs even if some part of the pipeline tweaks mode in-place; no more erratic quantization choices caused by reused mutable defaults.
• Since the API stays identical—still defaulting to ["relu", None]—existing call sites require no changes, just safer semantics.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/3367

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit bbf915a with merge base 53b5efd ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.