test: align dispatch tests with provider-scoped validation and named registration errors

Follow-ups for the salvaged wave: the auto-detect legacy-error test now
stubs the split validators, the unknown-command-provider test expects
the new provider_not_registered error, and _transcribe_local tolerates
a null stt.local config section again.

tests/tools/test_transcription_command_providers.py

@@ -528,7 +528,10 @@ class TestTranscribeAudioDispatchToCommandProvider:
         with patch("tools.transcription_tools._load_stt_config", return_value=cfg):
             result = transcribe_audio(str(audio))
         assert result["success"] is False
-        assert "No STT provider available" in result["error"]
+        # Explicitly-configured unknown providers now get a named
+        # registration error instead of the generic legacy message.
+        assert result["error_type"] == "provider_not_registered"
+        assert "unknown-cli" in result["error"]
 
 
 # ---------------------------------------------------------------------------

tests/tools/test_transcription_plugin_dispatch.py

@@ -256,6 +256,8 @@ class TestTranscribeAudioE2E:
         from unittest.mock import patch
 
         with patch("tools.transcription_tools._validate_audio_file", return_value=None), \
+             patch("tools.transcription_tools._validate_audio_source_file", return_value=None), \
+             patch("tools.transcription_tools._validate_audio_file_size", return_value=None), \
              patch("tools.transcription_tools._load_stt_config", return_value={}), \
              patch("tools.transcription_tools.is_stt_enabled", return_value=True), \
              patch("tools.transcription_tools._get_provider", return_value="none"):

tools/transcription_tools.py

@@ -1399,7 +1399,7 @@ def _transcribe_local(file_path: str, model_name: str) -> Dict[str, Any]:
             return {"success": False, "transcript": "", "error": "faster-whisper not installed"}
 
     try:
-        local_cfg = _load_stt_config().get("local", {})
+        local_cfg = _load_stt_config().get("local") or {}
         # Lazy-load the model (downloads on first use, ~150 MB for 'base').
         # Double-checked lock: concurrent voice messages must not both
         # download/load the model (#24767).