- Add chunk_transcription_mode (bool, default False) and overlap_time
  (int seconds, default 5) to settings.yaml and Pydantic AppSettings model
- RecordingSession gains chunk_transcription_mode, overlap_time,
  part_transcripts, prev_chunk_wav fields
- RecordingManager gains set_whisper_service() for background transcription
- When chunk_transcription_mode=True: after each chunk is written, it is
  immediately converted to WAV and transcribed via ThreadPoolExecutor
- Each chunk after the first prepends the last overlap_time seconds of the
  previous chunk's WAV audio (for context preservation)
- Transcript parts saved as part_000.txt, part_001.txt, etc.
- On stop: merge_chunks waits for all transcriptions and concatenates
  part_*.txt into transcript.txt; full_audio.wav is not created
- Normal mode (chunk_transcription_mode=False) unchanged
- New src/config/models.py with Pydantic settings models
- Updated /record/start and /record/stop routes to use new config
- Fix unused variable lint errors (transcript, transcript_dispatch)
- Fix test fixture: chunks_base_dir -> chunks_base

@@ -134,2 +145,4 @@
         raise HTTPException(status_code=400, detail="Cannot add chunk to inactive session")
 
+    # In chunk_transcription_mode, trigger immediate transcription of this chunk
+    if chunk_mode and result.chunk_count > 0:

@@ -173,2 +271,2 @@
-            transcript_path.write_text(transcription_text)
-            logger.debug(f"Saved transcript to {transcript_path}")
+    
+    if chunk_mode:

@@ -413,0 +435,4 @@
+            return None
+
+    def _transcribe_chunk_to_part_file(
+        self,

@@ -56,2 +55,4 @@
 # Request/Response models
+# =============================================================================
+
 class StartRecordingResponse(BaseModel):

@@ -243,4 +84,1 @@
-
-# --- Chunks management endpoints ---
-
 class ChunkFoldersResponse(BaseModel):

@@ -281,33 +119,289 @@ class SendToRepoResponse(BaseModel):
     error: Optional[str] = None
 
 
+class OutputConfig(BaseModel):

@@ -308,0 +207,4 @@
+        )
+
+
+class ChunkFolderTranscriptionMode:

@@ -308,0 +243,4 @@
+# Transcription helpers
+# =============================================================================
+
+def run_transcription(audio_path: Path, mode: TranscriptionMode) -> Optional[TranscriptionResult]:

@@ -515,0 +614,4 @@
+    return safe[:100].strip() or "structured"
+
+
+def _extract_headings_for_filename(md_content: str) -> tuple[str | None, str | None, str | None]:

@@ -515,0 +644,4 @@
+    return h1, h2, h3
+
+
+def _safe_filename_part(text: str) -> str:

@@ -515,0 +655,4 @@
+    h1: str | None,
+    h2: str | None,
+    h3: str | None,
+) -> str:

@@ -515,0 +667,4 @@
+    return "-".join(parts)
+
+
+def _build_structured_filename(folder_name: str, md_content: str) -> str: