2.7.3 — diagnostic: burst-size probe (POLL_PAD_TO)

⚠️ Diagnostic pre-release for #1. Not for general use.

Burst-size probe

Prior rounds ruled out sustained read rate as the crash trigger, and a crash appeared ~10 min after the cycle grew to ~10 reads (power + picture + device_info). This build separates the two remaining hypotheses — burst size vs a specific poison read.

New POLL_PAD_TO knob (const.py) pads each poll cycle up to N reads by repeating a known-safe read (power 124/1), adding no new distinct read.

Default config: power group only + POLL_PAD_TO = 10 → a 10-read burst of identical power queries per cycle.

• Crashes → it's burst / queue depth: too many queries in one tight cycle, regardless of content. Fix = cap/stagger reads per cycle (pairs with the planned poll-gating + tiered polling).
• Stays clean → burst count alone isn't it; the crash is content-specific → re-enable a group and sub-bisect.

How to run

1. Update via HACS (beta channel).
2. Set the poll interval to 5 s in Settings → Devices → Optoma Link → Configure.
3. Reboot HA. Only power is read (10× per cycle); all other entities stay "unknown".
4. Watch lamp hours for a crash. Set POLL_PAD_TO = 0 to disable padding.

See DIAGNOSTICS.md.