[akshitkhokhani]

mlx_whisper meeting.part1.wav currently writes meeting.txt: the CLI removes the audio extension, then the writer removes .part1 as another suffix. A separate invocation for meeting.part2.wav overwrites that transcript. Explicit --output-name draft.v2 similarly becomes draft.txt, despite the option describing a name before the output-format extension.

Append the format extension to the complete output name. Regression tests use real writers/files for TXT, VTT, SRT, TSV, JSON and all, plus separate single-input CLI invocations with only transcription mocked. They cover dotted audio basenames, an explicit dotted name, and unchanged plain names.

Validation:

• Original writer: 8 failing subtests; patched writer: both test methods and all subtests pass.
• Existing TestAudio: 3 tests pass on CPU using bundled audio.
• Repository Black/isort pre-commit hooks pass on both changed files; git diff --check passes.
• Model-dependent transcription tests were not run; no models were downloaded.

The separate multi-input CLI naming issue is tracked by #1441 / #1447. These tests use separate invocations to isolate the writer bug.

AI-assisted source investigation, implementation and tests; a separate AI reviewer checked the change. Human maintainer review is still required.