fix(logging): fix f-string logger calls and add missing boundary tests

[Kunal-Somani]

Summary

Two issues found across the service layer:

Issue 1 — f-string logger anti-pattern in AudioService and TranscriptService

Both services were using f-strings inside logger.error():

# Before (incorrect)
logger.error(f"[error] [Service Layer] [AudioService] ... {str(e)}")

Python's logging module performs lazy string interpolation only when %s placeholders are used. Passing an f-string means the string is always constructed, even when the ERROR level is disabled — wasting CPU on every exception path.

# After (correct)
logger.error(
    "[Service Layer] [AudioService] [extract_audio] An error occurred: %s",
    str(e)
)

This is consistent with the fix applied to AudioTranscriptionSentimentPipeline in PR #27.

Issue 2 — Three missing boundary tests in TestAudioService

The original test suite left three edge cases completely untested:

1. start_time_ms=0
The guard clause rejects start_time_ms < 0, making 0 the valid lower boundary. No test verified that 0 passes through and produces a correct result.

2. start_time_ms as float
The validation explicitly uses isinstance(start_time_ms, (int, float)) allowing floats, but no test ever passed a float value through the full extract_audio flow.

3. user_id=None through extract_audio
The _save_audio unit tests cover the None branch internally, but extract_audio was never called with user_id=None end-to-end. This left the forwarding of None to _save_audio unverified.

Files Changed

• app/services/audio_service.py
• app/services/transcript_service.py
• tests/unit/test_services/test_audio_service.py