test(agent): cover a BLE mouse that wakes as a new node

[kohjunhao]

Summary

Regression coverage for the reconnect path behind #1610: a Bluetooth-direct mouse that sleeps and comes back under a new OS node is published again, including when it comes back while device I/O is paused. Nothing in the inventory needed fixing; reading the reporter's logs, the stuck state matches the pause that never resumed (#1281), fixed by #1323 in 0.8.11, and I have asked them to retest there. These tests pin the path so a regression in either half shows up in CI.

Changes

• openlogi-agent-core: two replay tests driving the real inventory watcher against the replay backend. a_bluetooth_mouse_that_wakes_as_a_new_node_is_published_again removes the node and adds a new one with I/O open. a_mouse_that_wakes_while_device_io_is_paused_is_published_on_resume does the same with I/O paused and asserts nothing is published until resume, then that the mouse is. The direct probe exchanges are factored into a helper shared with the existing replay test.

Both tests bite: with I/O never resumed the second one times out on an empty inventory (the reported symptom), and with I/O never paused its "not published while paused" assertion fails.

Testing

cargo fmt --all -- --check
cargo clippy -p openlogi-device -p openlogi-agent-core -p openlogi-hid --all-targets -- -D warnings
cargo test -p openlogi-device -p openlogi-agent-core -p openlogi-hid

The two new tests passed 25 repeat runs. Not runtime-tested on hardware; test-only change.

Refs #1610

[greptile-apps]

Confidence Score: 5/5

[Low risk] Test file refactors cassette data for a BLE mouse scenario.

The PR appears safe to merge.

Summary

This PR adds replay coverage for a Bluetooth-direct mouse that reconnects under a new node, both with device I/O available and while it is paused. The latest change also asserts that the paused watcher does not open the woken node.

_{Reviews (2) · Last reviewed commit: "test(agent): assert a paused gate never ..."}