[elainewang-db]

Summary

agentbricks deploy prints the memory and session stores it reconciled, but --output json left them out. A script that deploys and then needs the stores, to inspect or delete them, had to parse agent.toml or read the App's environment variables instead.

This adds two fields to the deploy --output json result:

"memory_store": "my-agent-a1b2c3-memory",
"session_store": "my-agent-a1b2c3-sessions",

They're the store names from agent.toml, the same values the human output shows. Each is null when the project declares no store, matching how trace_experiment_id is null without tracing. No existing field changes.

The Agent Bricks E2E suites (#706) will use them: a test reads the store names from deploy's output and registers memory stores delete / sessions stores delete for them, instead of reading agent.toml.

Also: disable CLI telemetry in unit tests

Every CLI command sends a usage record from a background thread, through a real client built from the developer's ambient Databricks auth. In unit tests run on a machine with a [DEFAULT] profile in ~/.databrickscfg, that meant real databricks auth token calls and telemetry POSTs to that workspace. Tests that mock subprocess.run also saw those calls: the 5 pipeline_test.py tests that count subprocess.run calls failed locally (they pass in CI, which has no credentials). An autouse fixture in tests/unit_tests/conftest.py now sets AGENTBRICKS_DISABLE_TELEMETRY=1; the telemetry tests already unset it where they need telemetry on.

Testing

• ruff check and ruff format --check pass.
• pytest tests/unit_tests: 1813 passed, 211 skipped, with a [DEFAULT] profile configured (the 5 pipeline_test.py failures are gone).

This pull request and its description were written by Isaac.