fix(z2m): revert raw target decoder to 9-byte stride / int8 id (v3.2.5)

[nickduvall921]

Summary

Reverts the legacy raw-bytes target decoder back to a 9-byte stride with a signed int8 id, mirroring upstream Z2M PR Koenkk/zigbee-herdsman-converters#12284 (merged 2026-05-23).

Background

• The Inovelli 0xFC32 reportTargetInfo records are 9 bytes each: x, y, z, dop as little-endian int16 + id as a signed int8.
• Inovelli's mmWave docs briefly described id as int16 (10-byte stride). Z2M implemented that in #11915 (2026-04-11), and we mirrored it in v3.2.4.
• Inovelli corrected the docs — id is int8 — and Z2M reverted via #12284. Math.floor(buf.length / 10) was 0 for 9-byte buffers, so the parsed mmwave_targets array was always [] under the broken 10-byte parse.
• This PR realigns our raw fallback with real device traffic.

Changes

• app.py — _process_target_data: stride 10 → 9, byte-need check offset+9 → offset+8, id parsed via new parse_signed_8 instead of parse_signed_16.
• utils.py — add parse_signed_8() helper (matches parse_signed_16 style + fault tolerance).
• tests/test_parsing.py — +7 cases for parse_signed_8 (0, 1, 127, −1, −128, offset read, missing key).
• config.yaml — version 3.2.4 → 3.2.5.
• CHANGELOG.md — [3.2.5] entry.

Impact

• This path is dormant on Z2M ≥ 2.9 (gated off whenever parsed mmwave_targets is present, per 3.2.2), so it only affects pre-2.9 raw-fallback users, who now get correct coordinates again instead of garbage on target License? #2+.
• The parsed-mmwave_targets path benefits directly from #12284: Z2M builds carrying that fix repopulate the array, so modern setups get live targets back.

Test plan

• pytest tests/ → 207 passed (200 + 7 new)
• app.py parses / imports the new helper cleanly

🤖 Generated with Claude Code