fix: allow text files with non-multimodal models (#5137)

TextFiles passed via input_files incorrectly triggered a 'Model does not
support multimodal input' error for non-vision-capable models. Text files
are now inlined as text content in the message instead of being rejected.

Changes:
- base_llm.py: Rewrite _process_message_files to distinguish text files
  from binary files; add _is_text_file helper
- llm.py: Apply same text-file inlining logic in both sync and async
  _process_message_files methods
- crew.py: Recognize text file MIME types as auto-injectable so they
  don't require the read_file tool
- task.py: Same text-file auto-injection logic in prompt method
- tests: Add 17 tests covering text file inlining, image rejection,
  mixed files, _is_text_file helper, and edge cases

Co-Authored-By: João <joao@crewai.com>