fix: promote None-initialized kwarg to list in forward_capture

[ramkrishs]

Description

forward_capture in auto_round/calibration/hooks.py raised an AttributeError
during calibration when an optional kwarg (e.g. a sparse-attention mask in MoE
models) was None on the first calibration sample but a real Tensor on a later
one.

Root cause: The initialization branch stores None directly when a kwarg's
value is None:

if data is None or key in state.model_context.shared_cache_keys:
    state.inputs[name][key] = data   # stores None

On the next sample, the append branch assumed the stored value was always a list
and called .append() or .extend() on it, raising:

AttributeError: 'NoneType' object has no attribute 'extend'

Fix: One guard before the append/extend calls promotes a stored None to an
empty list, covering both batch_size <= 1 (.append) and batch_size > 1
(.extend) paths.

Type of Change

• Bug fix

Related Issues

Fixes #1950

Testing

Added three regression tests in test/test_cpu/core/test_forward_capture_none_kwarg.py:

Test	What it covers
test_none_then_tensor_kwarg_batch_size_1	None → Tensor transition, batch_size=1 (.append path)
test_none_then_tensor_kwarg_batch_size_gt1	None → Tensor transition, batch_size>1 (.extend path)
test_normal_tensor_kwarg_unaffected	Kwargs that are always Tensors still accumulate correctly

All three pass locally against the fix.

Checklist Before Submitting

• My code has been tested locally.
• Documentation has been updated as needed.
• New or updated tests are included where applicable.
• The CUDA CI has passed. You can trigger it by commenting /azp run Unit-Test-CUDA-AutoRound.

[ramkrishs]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Commenter does not have sufficient privileges for PR 1979 in repo intel/auto-round

[n1ck-guo]

LGTM. Thanks for your effor.

[chensuyue]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).