feat: expose SHT40 ambient temperature and humidity

[g4bri3lDev]

Requires py-opendisplay 7.15.0 (manifest bumped), which added the SHT40 decoder and the tracker filtering this uses.

Based on feat/clean-port, since main is 124 commits behind and lacks the current beta tag.

Ambient temperature + humidity

Boards with an SHT40 — reTerminal E1001/E1002/E1003/E1004 and Sticky — broadcast the reading in the advertisement's dynamic block. The passive coordinator already receives it, so these entities need no connection and no polling.

They're primary entities: not diagnostic, enabled by default. Ambient room temperature is what a user expects to find on the device, unlike the MCU's own temperature.

The offset within the dynamic block is per-board and cannot be assumed — E1001/E1002/E1004 use 1 while the firmware default is 7 — so it comes from the device's own config and is captured per entity at setup. This matters: decoding at the wrong offset produces plausible-looking garbage rather than an obvious error (a real button byte of 0x28 decodes as -39.9 °C / 4.0 %RH).

Failed reads (FF FF FF) and never-written slots (00 00 00) both report unknown rather than a bogus value.

Chip temperature renamed

The existing temperature entity is the MCU's, and with device_class=TEMPERATURE and no translation_key it displays as plain "Temperature" — exactly the name the new ambient entity would take. Two entities called "Temperature" on one device.

It's now "Chip temperature". This is a translation_key change only: the entity key, and therefore the unique_id, stays "temperature", so existing entities keep their entity_id and history. It was already EntityCategory.DIAGNOSTIC and disabled by default, so nothing else changes.

Only strings.json and en.json are touched; the other languages come from the translation workflow.

Phantom button events

AdvertisementTracker now receives the configured button byte indices. Only the bytes a BinaryInputs packet claims are buttons; the rest belong to touch controllers and sensors, and every byte decodes into a valid-looking button report — so a refreshed SHT40 reading emitted phantom button_slot_changed events on every 30-second poll.

event.py filters by byte_index before firing an entity, so nothing was user-visible; the events were simply produced and discarded on every advertisement. This is hygiene, not a user-facing bugfix.

Verification

104 tests pass (94 existing + 10 new); ruff clean on the changed files.

New coverage: decode at the default and at a relocated offset, msd_data_start_byte=0 resolving to the default slot, both sentinels reporting unknown, the entity-category/enabled-by-default split between ambient and chip, the chip entity keeping its unique_id key, per-instance key uniqueness, and native_value before any advertisement arrives.

Verified end to end on a real E1003 earlier in this work: after adding the sensor packet to its config, the device reports 28.0 °C / 63.6 %RH.

Note for whoever lands the TouchTracker refactor

There's WIP that moves TouchTracker construction into OpenDisplayCoordinator.__init__ as a third positional parameter. This PR adds binary_inputs in that same slot. Whichever lands second should reconcile them — making one or both keyword-only is the obvious fix.