fix(tracing): group MLflow traces by session across workflows

[dimakis]

Summary

• Propagate parent session into nested workflows — all 5 workflow entry points (discovery, conversation, resume_after_approval, analysis, optimization) now use session_context(get_session_id()) instead of session_context(), so nested workflows inherit the parent conversation's session ID rather than creating a new one
• Tag traces in trace_with_uri decorator — after each traced function completes, mlflow.update_current_trace(tags={"mlflow.trace.session": session_id}) is called, ensuring tool and HA client traces are grouped with their parent session
• Add start_experiment_run to optimization workflow — previously log_param/log_metric calls were silently dropped because no MLflow run was active
• Send conversation_id from frontend — streamChat now passes activeSessionId as conversation_id in the request body, eliminating the need for server-side derivation via message hashing

Files Changed

File	Change
src/graph/workflows/discovery.py	Inherit parent session
src/graph/workflows/conversation.py	Inherit parent session (2 entry points)
src/graph/workflows/analysis.py	Inherit parent session
src/graph/workflows/optimization.py	Inherit parent session + add start_experiment_run
src/tracing/mlflow.py	Add _tag_trace_session() to trace_with_uri
ui/src/api/client/conversations.ts	Add optional conversationId param
ui/src/pages/chat/index.tsx	Pass activeSessionId to streamChat
tests/unit/test_tracing_context.py	3 new tests for session inheritance
tests/unit/test_tracing_mlflow.py	3 new tests for trace_with_uri tagging

Test plan

• 2370 backend unit tests pass
• 13 frontend streamChat SSE parser tests pass
• No linter errors
• Manual: verify traces in MLflow UI are grouped by session after a multi-turn conversation
• Manual: verify nested workflow (e.g. discovery triggered from conversation) shares parent session

Made with Cursor