Persist compact JSON freeze-frame on fault confirm

[mfaferek93]

On the transition to CONFIRMED, snapshot capture now also persists a compact JSON freeze-frame of the captured topic values to SQLite, keyed by fault_code and retained across clear_fault (unlike rosbag snapshots).

• Capture set is config-driven via the existing per-fault-code snapshot topic resolution; a fault code with no capture configuration writes no freeze-frame row (a lookup returns not-found).
• Rows survive clear_fault, startup reclassification, and a DB reopen; a genuine re-capture replaces the row.
• Note: timer-driven confirmations do not enqueue snapshot capture (existing capture-path scope), so they produce no freeze-frame either.

Tested: freeze-frame unit + storage suites on both backends and capture-pool end-to-end; full fault_manager suite passes.

Closes #476

[Copilot]

Pull request overview

Adds persistent “freeze-frame” capture support to the fault manager so that, on snapshot capture, a compact JSON object of captured topic values is stored per fault_code and retained across clear_fault and DB reopen (SQLite + in-memory), with accompanying unit tests.

Changes:

• Extend FaultStorage with store_freeze_frame() / get_freeze_frame() and implement them for SQLite and in-memory backends.
• Persist a new freeze_frames SQLite table and write freeze-frame JSON during SnapshotCapture::capture().
• Add unit tests covering freeze-frame storage semantics (replace-on-recapture, survives clear, persists across reopen) and an end-to-end capture test that validates JSON content.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/ros2_medkit_fault_manager/test/test_sqlite_storage.cpp	Adds SQLite freeze-frame storage tests (CRUD, replace, survive clear, persist reopen).
src/ros2_medkit_fault_manager/test/test_snapshot_capture.cpp	Adds end-to-end capture test that validates freeze-frame JSON content; adds “unconfigured/disabled” cases.
src/ros2_medkit_fault_manager/test/test_fault_manager.cpp	Adds in-memory freeze-frame storage behavior tests.
src/ros2_medkit_fault_manager/src/sqlite_fault_storage.cpp	Creates freeze_frames table and implements store/get queries.
src/ros2_medkit_fault_manager/src/snapshot_capture.cpp	Accumulates per-capture JSON dict and persists it as a freeze-frame keyed by fault_code.
src/ros2_medkit_fault_manager/src/fault_storage.cpp	Implements in-memory freeze-frame persistence.
src/ros2_medkit_fault_manager/include/ros2_medkit_fault_manager/sqlite_fault_storage.hpp	Adds freeze-frame API overrides to SQLite storage interface.
src/ros2_medkit_fault_manager/include/ros2_medkit_fault_manager/snapshot_capture.hpp	Updates capture helpers to also fill a freeze-frame JSON object.
src/ros2_medkit_fault_manager/include/ros2_medkit_fault_manager/fault_storage.hpp	Introduces FreezeFrameData and adds freeze-frame API to FaultStorage.
src/ros2_medkit_fault_manager/CMakeLists.txt	Adds std_msgs dependency for snapshot capture test publisher usage.

Comments suppressed due to low confidence (1)

src/ros2_medkit_fault_manager/src/snapshot_capture.cpp:330

• capture_topic_from_cache() holds cache_mutex_ while calling storage_->store_snapshot() and while parsing JSON for the freeze-frame. On the background-capture path this can block the subscription callback from updating the cache, increasing contention and latency. Copy the cached payload under the lock, then release the lock before SQLite I/O and JSON parsing.

  std::lock_guard<std::mutex> lock(cache_mutex_);

  auto it = message_cache_.find(topic);
  if (it == message_cache_.end() || it->second.data.empty()) {
    RCLCPP_DEBUG(node_->get_logger(), "No cached data for topic '%s'", topic.c_str());
    return false;
  }

  const auto & cached = it->second;

[bburda]

Additional findings outside the diff:

• README not updated: ros2_medkit_fault_manager/README.md documents the snapshot subsystem, including that snapshots are deleted on clear, but not the new freeze-frame captured on the same confirm and retained across clear_fault - the exact contrasting behavior that section already describes. Per the docs gate, add it.

• Two test-coverage gaps on the new code: the background_capture=true cache path (freeze_frame[topic] = json::parse(cached.data) with the raw-string fallback) is never exercised - the new e2e test uses background_capture=false only; and the empty-{} row case (a configured fault whose topics all fail to sample writes a {} frame, so get_freeze_frame() is non-null) is not asserted, though OnDemandCaptureHandlesNonExistentTopic already drives that path and could assert it in one line.

[mfaferek93]

Both outside-the-diff points addressed in 88f54a7: the fault_manager README now documents the freeze-frame (retained across clear, contrasted with snapshots), and the two test gaps are covered - BackgroundCaptureCachesFreezeFrame exercises the background cache path incl. the json parse branch, and OnDemandCaptureHandlesNonExistentTopic now asserts the {} row.