Add per-command timeouts for ROS graph collection, harden integration script, and add tests

[RN0H]

Motivation

• Prevent a single slow or hung ros2 subcommand from stalling the whole graph collection by using per-command timeouts.
• Make the integration test script more robust and debuggable in CI by adding retries, readiness checks, and clearer failure output.
• Add unit tests to assert timeout/error behavior for graph collection logic.

Description

• Introduce a graphCommandTimeout (default 10s) and a helper runGraphCommand that wraps each ros2 invocation with its own timeout and replace previous shared timeout usage in CollectSystemGraph.
• Improve error handling in CollectSystemGraph by returning errors when ros2 topic list or ros2 service list fail.
• Harden tests/integration.sh by using unique container names, stronger container readiness checks, retrying gim graph with JSON validation via jq, adding debug output on failure, silencing/remove-stale-container calls, and setting GIMBLE_AI_API_KEY= when running gim analyze.
• Add two new unit tests (graph_errors_test.go and graph_timeout_test.go) to validate error propagation and per-command timeout behavior in CollectSystemGraph.
• Update CI workflow .github/workflows/sanity.yml to install jq in the integration job so the script's JSON checks work.

Testing

• Ran unit tests in the ROS package with go test ./internal/ros, which executed TestCollectSystemGraph_UsesPerCommandTimeout and the error tests and they passed.
• Exercised the integration script ./tests/integration.sh (as used by the CI integration job) with jq available and the script completed successfully in CI.
• Verified make build and the gim graph ros2/gim analyze ros2 flow in the integration script path during the integration run and observed expected JSON output.

---

Codex Task