chore: add type_checking dependency group for complete mypy coverage

Create a new type_checking dependency group that includes:
- All type stubs (moved from dev group for better organization)
- Optional runtime dependencies needed for complete mypy coverage
  (streamlit, anthropic, databricks-sdk, fairscale, torchtune, etc.)

This reduces mypy errors from 73 to 59 by providing proper type information
for all optional features. The dev group is now cleaner with only actual
development tools.

To run mypy with full type coverage:
  uv run --group dev --group type_checking mypy

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

pyproject.toml

@@ -72,6 +72,13 @@ dev = [
     "black",
     "ruff",
     "mypy",
+    "pre-commit",
+    "ruamel.yaml", # needed for openapi generator
+]
+# Type checking dependencies - includes type stubs and optional runtime dependencies
+# needed for complete mypy coverage across all optional features
+type_checking = [
+    # Type stubs
     "types-requests",
     "types-setuptools",
     "types-jsonschema",
@@ -79,8 +86,26 @@ dev = [
     "types-psutil",
     "types-tqdm",
     "boto3-stubs[s3]",
-    "pre-commit",
-    "ruamel.yaml", # needed for openapi generator
+    # Optional runtime dependencies for type checking
+    "streamlit",  # For UI type checking
+    "streamlit-option-menu",  # For UI type checking
+    "pandas",  # For UI and data processing type checking
+    "anthropic",  # For Anthropic provider type checking
+    "databricks-sdk",  # For Databricks provider type checking
+    "fairscale",  # For model parallelism type checking
+    "torchtune",  # For fine-tuning type checking
+    "trl",  # For HuggingFace training type checking
+    "peft",  # For HuggingFace LoRA/QLoRA fine-tuning type checking
+    "datasets",  # For dataset loading type checking
+    "together",  # For Together AI provider type checking
+    "nest-asyncio",  # For library client type checking
+    "pymongo",  # For MongoDB kvstore type checking
+    "torchvision",  # For multimodal model type checking
+    "sqlite-vec",  # For SQLite vector store type checking
+    "faiss-cpu",  # For Faiss vector store type checking
+    "mcp",  # For Model Context Protocol tools type checking
+    "ollama",  # For Ollama provider type checking
+    # Note: lmformatenforcer is not available in PyPI, so it remains an import-not-found error
 ]
 # These are the dependencies required for running unit tests.
 unit = [