Add Mxbai(mixbread-ai) Embedding Model to Hub

[laxmareddyp]

Description of the change

Add 4 mixedbread-ai (mxbai) embedding model presets to KerasHub.

License: Apache-2.0

Presets

Preset	Architecture	Pooling	Normalization	Params
mxbai_embed_large_v1_en	BERT (24L)	CLS	None	335M
mxbai_embed_2d_large_v1_en	BERT (24L)	CLS	None	335M
mxbai_embed_xsmall_v1_en	BERT (6L)	Mean	None	24M
deepset_mxbai_embed_de_large_v1	XLM-RoBERTa (24L)	Mean	L2	560M

Changes

• tools/checkpoint_conversion/convert_mxbai_embed_checkpoints.py : Added conversion script for all 4 presets to use from_preset("hf://...") to leverage existing BERT/XLM-RoBERTa converters.
• keras_hub/src/utils/transformers/convert_bert.py: Bug fix — added max_sequence_length to convert_backbone_config(). Without this, BERT models with non-512 max_position_embeddings (e.g., xsmall-v1 with 4096) fail during weight conversion due to position embedding shape mismatch.

Checklist

• I have added all the necessary unit tests for my change.
• I have verified that my change does not break existing code and works with all backends (TensorFlow, JAX, and PyTorch).
• My PR is based on the latest changes of the main branch (if unsure, rebase the code).
• I have followed the Keras Hub Model contribution guidelines in making these changes.
• I have followed the Keras Hub API design guidelines in making these changes.
• I have signed the Contributor License Agreement.

[gemini-code-assist]

Code Review

This pull request introduces a checkpoint conversion script for HuggingFace mixedbread-ai mxbai-embed checkpoints to KerasHub format, supporting both BERT and XLM-RoBERTa architectures, and updates the BERT config converter to map the maximum sequence length. The review feedback recommends enhancing the conversion script's robustness by raising errors if numerical verification or semantic search ranking fails, rather than just printing diagnostics, to prevent saving corrupted presets.