Fix bugs related to lightning 2.6.0 and transformer 5.0.0

[chcwww]

What does this PR do?

This PR aims to fix two bugs introduced by the lightning 2.6.0 and transformer 5.0.0.

1. In the release of lightning 2.6.0 (link), they exposed weights_only argument and let torch handle default value, which was set to True in torch 2.6.
   However, before this update, lightning used weights_only=False as the default for torch.load().
   Therefore, after the update, we have to explicitly add weights_only=False to avoid registration problems during serialization.

2. In the release of transformer 5.0.0 (link), they no longer handle torchscript in the __init__() of class PretrainedConfig (which they did in previous versions: link).
   Therefore, this flag is no longer removed during the initialization of the config class (link), and ends up being passed to class BertModel directly as unused_kwargs.
   However, torchscript is not a valid argument for class BertModel.
   To sum up, we decided to remove torchscript=True.

Test CLI & API (bash tests/autotest.sh)

Test APIs used by main.py.

• Test Pass
  • (Copy and paste the last outputted line here.)
• Not Applicable (i.e., the PR does not include API changes.)

Check API Document

If any new APIs are added, please check if the description of the APIs is added to API document.

• API document is updated (linear, nn)
• Not Applicable (i.e., the PR does not include API changes.)

Test quickstart & API (bash tests/docs/test_changed_document.sh)

If any APIs in quickstarts or tutorials are modified, please run this test to check if the current examples can run correctly after the modified APIs are released.

[Eleven1Liu]

LGTM