release: v0.5.0

[zebraengine]

Problem

PyPI serves 0.4.0; main carries two features a new install should have from day one: a verified, compressed backup command (#26) and per-install calibration of the idle-offset model (#27). Without a release, uvx wallmonitor still ships the one-install offset constants as the only model.

Code touched

pyproject.toml — 0.4.0 → 0.5.0; uv.lock refreshed to match. Nothing else; the release content is what already merged:

• feat: one-shot verified, compressed database backup into any directory #26 — wallmonitor --backup DIR: online snapshot, integrity-checked, gzip/xz, atomic placement, date-based rotation; opt-in installer --backup-dir timer. Nothing scheduled by default.
• feat: calibrate the idle-offset model per install from its own sensor history #27 — idle-offset model calibrated from the install's own stationary-sensor history in the daily maintenance pass, adopted on material sane change, event-marked; sensor-less installs keep the labelled seed with a stated ±1.5 °C. New /api/thermal fields: model.idle_offset, forecast.ambient_se_c, idle ambient_se_c; new event kind idle_offset_calibrated.

Risk

Minor bump: new CLI surface and additive API fields; no schema migration (one new settings key is created on first adoption). Installs with a stationary sensor will see one idle_offset_calibrated event on first start after upgrade and a small reinterpretation of proxy-tier fits (+0.05 to +0.4 °C on the reference install — inside the drift watch's noise floor). The daily maintenance pass now runs a ~5 s calibration scan on a month of raw rows at startup and every 24 h.

Verification

127/127 on main; each included PR carried its own verification, including read-only replays and a live adoption on the reference install. Tag push triggers the trusted-publishing workflow; after publish: PyPI JSON and uvx --from wallmonitor==0.5.0 wallmonitor --help.

Deploy: tag v0.5.0 on the merge commit after merge; GitHub Release created published. The reference install already runs this code.