Support known speaker hints in REST API

[boxerab]

Summary

• support OpenAI-compatible REST known speaker fields instead of warning that they are ignored
• allow multipart known_speaker_references uploads to enroll named speakers with the existing diarizer
• include speaker labels on verbose_json segments when REST speaker hints are provided
• document REST known speaker fields and add diarizer/helper coverage

Tests

• ruff check README.md tests/test_diarization.py tests/test_server_extended.py whisper_live/diarization.py whisper_live/server.py
• python -m py_compile whisper_live/diarization.py whisper_live/server.py tests/test_diarization.py tests/test_server_extended.py
• PYTHONPATH="$tmpdir:$PWD" pytest tests/test_diarization.py tests/test_server_extended.py -q with a temporary local torch stub because this workstation does not have torch installed on system python: 69 passed

Notes

• Full ruff check . still reports pre-existing lint issues outside this change.
• black --check on the touched files would reformat unrelated existing code in server.py and tests/test_server_extended.py, so this PR avoids that formatting churn.

[boxerab]

As all tests are passing and this is needed by Aavaaz project, going to merge.