fix: make /api/settings/status fast and truthful

[luca-71]

The endpoint hung long enough to look broken, leaving the settings panel unable to load — and once it did answer, one of its three answers was wrong.

1. No overall time budget

Each individual osascript call was already bounded, but the endpoint total was not. Probing Calendar went through get_todays_events(), which on a cold cache fans out over every calendar in batches of two at 15s apiece. On a 10-calendar account that alone runs to 75s, before Mail (20s) and Notes (15s) are even reached. Prior to the user granting TCC permission, these block on a system dialog instead.

Added _probe_integration(), capping each probe at 8s and reporting a slow integration the same as a broken one, and run the three concurrently via asyncio.gather — they touch independent apps, so the endpoint now costs one budget rather than three.

2. The probes could not see failure

calendar_access and notes_access both answer an unreachable app with the same empty value they use for "nothing found". A probe that only caught raised exceptions therefore reported an app as healthy while its script was failing on every single call — which is exactly what was happening to Notes.

Calendar and Notes are now probed by listing containers (get_calendar_names() / get_note_folders()), with an empty result read as failure. Listing also sidesteps the expensive paths entirely.

This part matters on its own: capping the wait alone would have left the endpoint honest about its deadline but wrong about its result.

3. One unreadable note emptied the whole listing

get_recent_notes() read note properties unguarded, so a single note that Notes.app refuses to describe aborted the entire AppleScript and the function returned [].

Each note is now guarded individually, and the scan continues until count readable notes are collected — bounded, so a run of bad notes cannot walk an entire library.

The folder lookup gets its own guard. name of container of note turns out to fail with -1728 for every note on recent macOS: the container resolves, but as a bare item whose name is not exposed. So folder is empty in practice. Real folder names are only reachable by walking every note of folder, which scans the whole library on every call — not worth it while no caller reads the field. Guarding rather than dropping the lookup keeps it self-healing if Apple restores the property.

Measured

	before	after
Response time	hung (>60s)	0.33s warm / 2.3s cold
calendar_accessible	never (timeout)	true
notes_accessible	true (while failing on every call)	true (and actually working)
get_recent_notes(10)	[]	10 notes

Verified against a 10-calendar, 140-note account. The recurring Notes script failed: ... (-1728) warnings are gone from the server log.

Tests

pytest tests/ gives 35 passed, 8 failed — identical before and after these changes, so no regressions. The 8 pre-existing failures are unrelated: 7 need playwright install, and test_browse_action_keywords imports ACTION_KEYWORDS, a symbol that no longer exists in server.py.

🤖 Generated with Claude Code