[regression] qkv_scale is empty due to offload in autoscheme.

[xin3he]

Description

Please briefly describe your main changes, the motivation.

Type of Change

• Bug fix
• New feature
• Documentation update
• Performance improvement
• Code refactoring
• Other (please specify):

Related Issues

Fixes or relates to #1573, #1574

Checklist Before Submitting

• My code has been tested locally.
• Documentation has been updated as needed.
• New or updated tests are included where applicable.

[Copilot]

Pull request overview

Fixes a regression where KV-cache scale parameters (e.g., k_scale/v_scale) can end up empty/zero when AutoScheme runs with offloading enabled, by adjusting how parameters are updated and adding a targeted CPU regression test.

Changes:

• Adds a CPU regression test covering AutoScheme + static_kv_dtype="fp8" to ensure k_scale/v_scale are populated.
• Updates update_parameter_data to handle cases where a parameter was cleared to an empty tensor (e.g., via offload) by recreating the parameter when shapes differ.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 5 comments.

File	Description
test/test_cpu/schemes/test_auto_scheme.py	Adds a regression test intended to detect missing/zero KV scales after quantize_and_save.
auto_round/experimental/utils.py	Modifies parameter update logic to try to recover when a parameter tensor has been cleared/shape-changed due to offload.

[chensuyue]

Still got the issue, #1573

[chensuyue]

Use #1583 instead.