fix: Docker paths, chunker word→char split, and ChatOrchestrator args

Fixed multiple issues preventing the indexer from running:
- Docker COPY paths: companion/ → src/companion/ to match project structure
- pyproject.toml: [tool.hatchling] → [tool.hatch.build.targets.wheel]
- api.py: ChatOrchestrator init params (session_memory instead of http_client)
- chunker.py: Fixed character-based chunking (was word-based, causing 400 errors from Ollama embedding API due to exceeding token limits)
- config.json: Use exact model tag mxbai-embed-large:335m
- docker-compose.yml: Fixed vault mount path

config.json

@@ -55,7 +55,7 @@
   "rag": {
     "embedding": {
       "provider": "ollama",
-      "model": "mxbai-embed-large",
+      "model": "mxbai-embed-large:335m",
       "base_url": "http://localhost:11434",
       "dimensions": 1024,
       "batch_size": 32