Expose BERT token-type embeddings through TransformerBridge

[LarryHu0217]

Description

Expose BERT token-type (segment) embeddings as the bridge-native token_type_embed.hook_out hook.

The BERT adapter now maps bert.embeddings.token_type_embeddings through an EmbeddingBridge. The regression test boots bert-base-cased, captures the new hook with run_with_cache, and compares it directly with Hugging Face's token-type embedding output.

Fixes #1662

Type of change

• New feature (non-breaking change which adds functionality)

Validation

• uv run pytest tests/unit/model_bridge/supported_architectures/test_bert_adapter.py -q --tb=short -o addopts=: 22 passed
• make unit-test: 5,260 passed, 56 skipped, 54 deselected, 10 xfailed
• make check-format: passed
• uv run mypy .: no issues in 431 source files
• make docstring-test: 17 passed, 24 skipped
• make acceptance-test: 199 passed, 72 skipped, 11 deselected
• make integration-test: 1,016 passed; the two google/gemma-2-2b-it cases could not run locally because this machine is not authenticated for that gated Hugging Face model (401)

Checklist

• The change is self-explanatory and does not need additional inline comments
• No documentation change is needed for this bridge-native hook mapping
• My changes generate no new warnings
• I have added tests that prove the feature works
• New and existing unit tests pass locally with my changes
• I have not rewritten tests relating to key interfaces which would affect backward compatibility

[jlarson4]

Looks great, thanks @LarryHu0217!