Auto-recover a hung preamp on persistent I2C write failures

[stamateviorel]

What does this change intend to accomplish?

When the preamp microcontroller hangs it stops ACKing and every I2C write fails with OSError 121 (EREMOTEIO). The existing fallback in _Preamps.write_byte_data only reopens the SMBus handle, which recovers a transient bus glitch but not a hung preamp — zone control stays dead until someone power-cycles the unit. This escalates the fallback: when the reopened-bus retry also fails, reset the preamps in place, re-assign I2C addresses, reopen the bus, re-flush all cached register values (so zone mute/source/volume state survives the reset), then retry the write. Rate-limited to once per 20 s so a benign one-off glitch never resets audio.

We hit this live on 2026-06-04 (all zone control dead, reads OK / writes failing with Errno 121, only a reboot helped). With this patch deployed the same wedge self-heals in under a second.

Checklist

• Have you tested your changes and ensured they work? (in production on a real AmpliPi since 2026-06-04)
• Have you checked to ensure there aren't other open Pull Requests for the same update/change?
• If applicable, have you updated the CHANGELOG?
• Does your submission pass linting & tests? (python -m py_compile clean; happy to fix anything CI flags)

[linknum23]

We probably need to merge this code with recent changes made with a similar intent.

[stamateviorel]

Makes sense — I dug into ResetOnFailure and I think these two layer together nicely rather than overlap.

Your branch detects the failure and surfaces it to the user (the failure counter + alert pointing at Settings -> Config -> Hardware Reset), which is exactly the right UX when the hardware genuinely needs attention. This PR sits one level below that: on the failure it automatically pulses reset_preamps() and re-flushes the cached register state so zone mute/source/volume survive, then retries — so the common transient wedge self-heals in well under a second without anyone having to touch the unit. Same reset_preamps() primitive your manual Hardware Reset uses, just triggered automatically.

The clean combination: auto-recover first; if the auto-recovery itself keeps failing, that's when your alert should fire — i.e. "we tried to self-heal and couldn't, this one really does need hardware attention." That's a stronger signal than alerting on the first I2C error, which is often transient.

Happy to rebase this onto ResetOnFailure so it merges into the same write_byte_data path with no conflict, and wire the escalation into your add_alert flow. Want me to put up that combined version?