Remove max_prompt_length from experimental BCO

[albertvillanova]

Remove max_prompt_length from experimental BCO.

This PR removes the max_prompt_length parameter from the BCO training pipeline, simplifying both the configuration and the tokenization logic.

Follow-up to:

• Remove max_prompt_length from experimental PRM #4963
• Refactor KTO coordinated with DPO [b/N]: Simplify truncation logic #4808
• Deprecate max_prompt_length in RLOOTrainer #4703
• Refactor DPO #3906

Configuration and Argument Cleanup

• Removed the max_prompt_length argument from the BCOConfig class and its documentation, as well as from all related code paths and warnings. This reduces user confusion and simplifies configuration.

Tokenization and Preprocessing Logic

• Updated the _process_tokens function to eliminate prompt truncation based on max_prompt_length. Now, only the response is truncated if the total sequence exceeds max_length, and prompt tokenization no longer uses a prompt length limit.
• Removed the passing of max_prompt_length in dataset mapping and processing, ensuring consistent use of the new logic across training and evaluation.

Example and Script Updates

• Cleaned up example scripts by removing --max_prompt_length from CLI arguments, aligning usage examples with the new configuration.

Testing updates:

• Removed all uses of max_prompt_length from test cases.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[qgallouedec]

yes looks good