perf(gui): clear DeviceList in one pass and fix discovery timer lifecycle

[geraldo-netto]

What

Three focused commits on blueman/gui/DeviceList.py, each building on the previous:

1. fix: track and cancel the discovery progress timer — discover_devices started a GLib.timeout for update_progress but discarded the source id, so stop_discovery() left it running until the callback next observed discovering=False. Store the id, remove it in stop_discovery() via _remove_discovery_timeout, and null it on the callback's own exit paths so a self-ending timer is never double-removed.
2. perf: clear in one pass instead of per-row deletion — clear() looped over the liststore calling device_remove_event per row. Each GenericList.delete() calls the O(n) Gtk.ListStore.iter_is_valid, making the whole clear O(n²), and it mutated the liststore while iterating it before calling liststore.clear() anyway. Replaced with a single liststore.clear() — fixes the iterate-while-mutating bug and drops the redundant pass.
3. perf: release row references before clearing the store — clearing while path_to_row still held a live Gtk.TreeRowReference per row forced GTK to update every reference as rows were removed (still ~O(n²)). Reset path_to_row before clear() so there are no live references to keep in sync.

Why

On adapter switch, populate_devices, and adapter removal, clear() runs over the full device list. With many paired/seen devices it was quadratic and relied on a buggy mutate-while-iterating loop.

Proof

test/benchmarks/bench_devicelist_clear.py exercises the original, step-2, and step-3 strategies on a real Gtk.ListStore with a matching TreeRowReference cache:

rows	original	one-pass clear (live refs)	release-then-clear
4000	162 ms	231 ms	37 ms

The decisive win is releasing references before clearing (~4.3× vs original at 4000 rows, near-linear vs quadratic). Note that a single clear() with the references still alive is not faster on its own — GTK keeps every live reference in sync per removal — which is why step 3 is a distinct change rather than a duplicate of step 2.

Tests

test/gui/test_devicelist.py (13 cases): discovery-timer store / cancel / null paths, the no-double-remove completion case, and clear() on empty/non-empty stores, reference release, and a size fuzz (0…1000). ruff + flake8 clean.

Note on behaviour

clear() keeps the same observable result (empty store, empty cache, one device-selected(None, None) emit when non-empty); the per-row device-selected emits the old loop produced for the selected row were redundant with the final emit.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud