v0.0.22

Added

• Hook tests (23 new tests). The four data hooks are now covered by real
  behavioral tests, rendered through a minimal in-repo hook harness
  (test/domSetup.ts: happy-dom document + react-dom root under act() — no
  testing-library dependency):
  • usePolledData — initial load, self-scheduled re-polls, render stability
    on unchanged payloads (same data reference), error set + clear on recovery,
    the generation guard dropping a stale in-flight result after a deps change,
    refetch(), and unmount stopping the loop.
  • useActivityStream — driven end-to-end through the real sse.ts parser
    via a mocked fetch stream: handshake → connected, newest-first ordering
    through the 150 ms flush, status/counter mapping, the 250-event ring cap,
    and the 2 s reconnect after a clean stream end.
  • useThroughputSeries — immediate first sample (series + latest + summed
    depth), swallowed transient failures; plus pure depthTrend cases
    (draining/accumulating/steady, dead band).
  • useAlertEngine — breach + single toast, edge-triggered no-re-notify, the
    skip-tick-when-overview-is-down policy (no false <-rule trips), per-source
    null metrics on partial failure, queue-scoped and p99_latency resolution,
    and the idle no-rules path (zero polling).
• happy-dom (dev-only) provides the DOM; the published package stays
  zero-dependency.

Changed

• Coverage floors raised to lines ≥ 70% / functions ≥ 58% (from 62%/55%),
  matching the new measured totals (72.7% / 60.3%).

Fixed

• Release publish race. release.yml's asset globs overlapped
  (bunqueue-dashboard-*.zip and bunqueue-dashboard-v* both matched the
  zip), so the same asset was uploaded twice concurrently — the race broke the
  first v0.0.22 publish (draft left without the zip) and duplicated the zip
  line in every release's SHA256SUMS. The globs are now disjoint
  (bunqueue-dashboard-v*-* matches only the platform binaries).

Full Changelog: v0.0.21...v0.0.22