feat: add control plane event tracking

[datasciencemonkey]

Summary

• Adds SDK User-Agent based control plane event tracking so CoDA usage appears on the shared Labs usage dashboard
• Follows the DQX pattern: piggybacks on the Databricks SDK's User-Agent header, transmits via clusters.select_spark_version()
• All events fire in background daemon threads — never blocks Flask request handling or terminal I/O
• Fire-and-forget: errors are caught and logged, never raised

Control Plane Events

Active Events (explicit log_telemetry() calls)

Event	Key	Value	Location	When it fires
App startup	event	app_startup	app.py:initialize_app()	Once per app start/restart
Session creation	agent	claude|codex|gemini|opencode|hermes|shell	app.py:create_session()	Every new terminal pane
File upload	event	file_upload	app.py:upload_file()	Every clipboard paste / file upload
PAT rotation	event	pat_rotation	pat_rotator.py:_rotate_once()	Every 10 minutes (while sessions exist)
Workspace sync	event	workspace_sync	sync_to_workspace.py:sync_project()	Every git commit inside ~/projects/

Passive Events (product_info on existing SDK calls)

All existing WorkspaceClient instances now carry _product_info = ('coda', version), so regular SDK calls (identity lookups, Apps API, workspace sync) also identify CoDA in the User-Agent header:

• app.py:_setup_git_config() — git identity lookup
• app.py:get_token_owner() — owner resolution (Apps API + PAT fallback)
• sync_to_workspace.py:get_user_email() — workspace sync user lookup

Files Changed

File	Change
telemetry.py	New — log_telemetry(), @telemetry_logger(), set_product_info()
tests/test_telemetry.py	New — 17 tests covering all telemetry functions
app.py	Import + 3 integration points + set_product_info on 3 WS clients
pat_rotator.py	log_telemetry call after successful rotation
sync_to_workspace.py	log_telemetry call after successful sync + set_product_info

Design Decisions

• Background threads instead of synchronous (DQX pattern is sync) — CoDA is a long-running web server, not a batch Spark job
• No env vars for WS client — reads from ~/.databrickscfg which pat_rotator.py keeps updated
• No DataFrame-level tracking — CoDA is a terminal app, not a data quality engine
• Decorator adapted — DQX's @telemetry_logger expects self.ws on class methods; CoDA's version works on standalone functions by creating its own WS client

Test plan

• 17 new tests in tests/test_telemetry.py — all passing
• Full test suite: 215 passed, 1 failed (pre-existing test isolation issue in test_session_detach.py, unrelated)