Support max_seq_len argument in multimodal text-to-text export

[kamalkraj]

Summary
This PR updates the load_multimodal_text_to_text_model function to support the max_seq_len argument, allowing users to explicitly define
the maximum sequence length during model export.

Changes

• Modified optimum/exporters/executorch/tasks/multimodal_text_to_text.py.
• Added logic to check for max_seq_len in kwargs.
• Sets max_length to max_seq_len if provided; otherwise, it falls back to the existing max_length logic (defaulting to 2048).

Motivation
This change improves flexibility by accommodating the max_seq_len parameter, which is often used in ExecuTorch configurations, while
maintaining backward compatibility with max_length.