fix: make shared breaker phase mapping configurable

[analienx]

Regression fix for zigbee2mqtt/hassio-zigbee2mqtt#871

The static TS0601 / _TZE204_wbhaespm identity is shared by physically different breaker variants, so the dedicated priority SMTONOFF matcher introduced by Koenkk#12972 cannot safely infer the physical product or phase order.

This RC2 is forward-ported directly onto current upstream 26.108.1 (22a4c98f66b60fcc9886334cae47751acf47ec79) and keeps the collision family on the shared definition while making DP6/DP7/DP8 phase order an explicit per-device option.

Behavior

• default phase_mapping: abc: DP6=A, DP7=B, DP8=C; preserves historical shared-family behavior and fixes the Add support for Drayton Wiser Smart TRV Koenkk/zigbee-herdsman-converters#871 regression
• explicit phase_mapping: cba: DP6=C, DP7=B, DP8=A; preserves the mapping physically verified on the tested SMTONOFF ZXB3-125 three-phase units
• static A/B/C exposes in either mode
• no runtime phase-count inference
• no dynamic exposes or new static fingerprint assumptions

Evidence

The earlier RC1 implementation was physically canary-tested on both installed SMTONOFF ZXB3-125 3P breakers with phase_mapping: cba: the setting persisted across Zigbee2MQTT restart, live phase telemetry remained correct, Home Assistant entities/statistics remained intact, and no unrelated device was captured by the shared definition.

This RC2 carries the same logical fix onto current upstream after tuya.ts changed.

RC2 freeze

Exact SHA: b278d857df8b8d12888e93f5ed9cb8ea4213ed65

Base: upstream 26.108.1 / 22a4c98f66b60fcc9886334cae47751acf47ec79

Scope: exactly one commit and three files:

• src/devices/tuya.ts
• src/lib/exposes.ts
• test/smtonoff-zxb3.test.ts

Fresh PR CI on the exact RC2 SHA is fully green:

• Build ✅
• Check ✅
• full Tests ✅ (28 files / 849 tests, including four focused SMTONOFF regression tests)
• Bench ✅
• autofix ✅

The dedicated ZXB3-125 priority matcher is removed. The shared _TZE204_wbhaespm definition gets the selectable phase mapping with abc as the compatibility-first default.

User configuration

Zigbee2MQTT: Device → Settings (specific) → Phase mapping.

• abc = DP6=A, DP7=B, DP8=C
• cba = DP6=C, DP7=B, DP8=A

Equivalent persisted device configuration:

devices:
  '<ieee_address>':
    phase_mapping: cba

This is the current upstream-ready candidate. RC1 in #6 is retained only as the physically tested predecessor artifact.

[analienx]

Closing the fork-local mirror because the exact RC2 head b278d857df8b8d12888e93f5ed9cb8ea4213ed65 is already the head of the real upstream PR Koenkk#13156. Keeping both PRs open adds no execution or review value; the fork branch remains the source branch for the upstream PR.