Refactor STT pipeline and CLI documentation

Split the STT worker into a collector and a transcription worker
to offload heavy processing to a background thread. Add the
`--whisper-model` flag and implement LLM latency logging. Expand
the README with comprehensive CLI usage instructions.

main.py

@@ -53,6 +53,14 @@ def main():
         help="Base URL for the LLM backend",
     )
 
+    # STT Configuration Arguments
+    parser.add_argument(
+        "--whisper-model",
+        type=str,
+        default=os.environ.get("WHISPER_MODEL", "base"),
+        help="The Whisper model to use for STT",
+    )
+
     # Pipeline Execution Argument
     parser.add_argument(
         "--run-pipeline",
@@ -75,7 +83,7 @@ def main():
     if args.run_pipeline:
         async def run_pipeline():
             loop = asyncio.get_event_loop()
-            orchestrator = PipelineOrchestrator(loop, llm_config=llm_config)
+            orchestrator = PipelineOrchestrator(loop, llm_config=llm_config, whisper_model=args.whisper_model)
             try:
                 await orchestrator.run()
             except KeyboardInterrupt: