fix(speech): use correct file extension for default output filename

[fifteen42]

Problem

When --format is specified without --out, the default output filename always used .mp3 regardless of the requested format:

mmx speech synthesize --text "Hello" --format wav
# Produces: speech_2026-04-09-07-15-14.mp3
# Actual content: RIFF WAVE audio

This means the file extension does not match the actual audio format. Tools that rely on file extensions (ffmpeg, audio editors, upload APIs, scripts) will misidentify or fail to process the file.

Fix

Derive the extension from the --format flag, defaulting to mp3 when not specified:

// Before
const outPath = (flags.out as string | undefined) ?? `speech_${ts}.mp3`;

// After
const ext = (flags.format as string) || 'mp3';
const outPath = (flags.out as string | undefined) ?? `speech_${ts}.${ext}`;

This only affects the auto-generated filename. Users who explicitly pass --out <path> are unaffected.

Note: music/generate.ts already handles this correctly — this brings speech/synthesize.ts in line with the same pattern.

Reproduction

mmx speech synthesize --text "hi" --format wav
file speech_*.mp3   # RIFF WAVE audio — extension mismatch

After fix:

mmx speech synthesize --text "hi" --format wav
# Produces: speech_xxx.wav  ✓