[DRAFT] Feat (axe): improved implementation with extended support/testing

[i-colbert]

Reason for this PR

The initial implementation of accumulator-aware extensions (AXE) for post-training quantization only supports asymmetric activation quantization with per-row scaling factors at the finest granularity, although the algorithm is general enough to support symmetric quantization with fine-grained scaling factors (i.e., per-group). This PR extends support to include both symmetric activation quantization and groupwise scaling factors, which enables AXE for MX datatypes.

The implementation follows the paper "Accumulator-Aware Post-Training Quantization" (see https://arxiv.org/abs/2409.17092)

Changes Made in this PR

An AXE mixin is created (i.e., AXE), which is compatible with both A2GPTQ and A2GPFQ. The mixin handles both the soft constraint that derives the optimal Lagrange multiplier and the hard constraint which handles the greedy recursive L1 bound.

The APIs for the TorchVision and Huggingface models remain unchanged; the backend has been simplified.

Blocked by #1172

Testing Summary

With the accumulator-aware GPxQ variants moved into the core library, the GPxQ unit tests have been extended.

Risk Highlight

• This PR includes code from another work (please detail).
• This PR contains API-breaking changes.
• This PR depends on work in another PR (please provide links/details).
• This PR introduces new dependencies (blocked by 1172).
• There are coverage gaps not covered by tests.
• Documentation updates required in subsequent PR.

Checklist

• Code comments added to any hard-to-understand areas, if applicable.
• Changes generate no new warnings.
• Updated any relevant tests, if applicable.
• No conflicts with destination dev branch.
• I reviewed my own code changes.
• Initial CI/CD passing.
• 1+ reviews given, and any review issues addressed and approved.
• Post-review full CI/CD passing.