Initial commit: sermon-clean v0.1.0

Find, cut, paste replacements in sermon audio.
- engine.py: SermonClean core (BadSegment, trim, concat, splice verify)
- cli.py: find/cut/paste/pipe subcommands
- elevenlabs.py: optional ElevenLabs rendering hook
- 25 tests passing (timestamp parsing, BadSegment validation, end-to-end trim on real audio)
- pyproject.toml: pip-installable; dep only on ffmpeg/ffprobe system-wide
- README + LICENSE (MIT) + examples/segs.json

Validated against the krystie audio cache (OGG/Opus 48kHz mono).

examples/segs.json

@@ -0,0 +1,14 @@
+[
+  {
+    "start": 1298.0,
+    "end": 1302.0,
+    "reason": "misspoke",
+    "replacement_text": "the actual sentence you meant to say"
+  },
+  {
+    "start": 1450.3,
+    "end": 1453.1,
+    "reason": "misspoke",
+    "replacement_text": "the corrected phrase"
+  }
+]