Fix PT2E Mul weight detection for parameters

[ricardoasouz]

Summary

Fixes PT2E/TorchAO weight detection for aten.mul.Tensor when a learned parameter can appear on either operand.

Previously, TorchAOQuantizerAdapter relied on metatype.weight_port_ids to determine whether a get_attr node represented a weight. PTMulMetatype has no fixed weight port because both

• activation * weight
• weight * activation

are valid.

As a result, a torch.nn.Parameter feeding Mul could be classified as an activation.

This change adds a Mul-specific fallback: when the source get_attr corresponds to a named torch.nn.Parameter, it is treated as a weight. Other operations continue to use the existing weight_port_ids behavior.

Tests

Added coverage for:

• activation * Parameter -> WeightQuantizationInsertionPoint
• Parameter * activation -> WeightQuantizationInsertionPoint
• activation * buffer -> remains ActivationQuantizationInsertionPoint
• custom TorchAO quantizer with batch size 2
• automatic batchwise_statistics=True
• explicit verification that quantization is not silently skipped
• explicit Q/DQ presence in the resulting FX graph

Validation:

• focused issue coverage: 4 passed
• ViT regression checks: 2 passed
• tests/executorch/test_ptq.py: 57 passed
• make pre-commit: passed
• git diff --check: passed

Notes

The behavior change is intentionally limited to PTMulMetatype so parameters used by other operations continue to follow their existing metatype-defined weight-port semantics.

Related to #3576.

AI-assisted development

AI tools were used during codebase exploration, hypothesis generation, and drafting. I personally reviewed, tested, and validated the final implementation and can explain and maintain the submitted changes.