fix: config serialization correctness — padding, mfr metadata, sizes, tx_power (C7/C8/M5/M6)

[balloob]

Summary

Four related config wire-format correctness fixes.

C8 — pad fixed-size buffers on serialize (🔴)

Every serializer except display did reserved[:N] with no padding. The TLV format has no per-packet length — the firmware memcpys a fixed sizeof(struct) — so one short packet shifts and misparses everything after it on the device. All fixed-size reserved buffers are now ljust-padded to their exact width.

C7 — ManufacturerData round-trip corrupted bytes 4–21 (🔴)

_parse_manufacturer_data stored data[4:22] into reserved and left the simple-config fields at 0, but the serializer writes simple_config_driver/display/power_index + configured_at at offsets 4–15. A read-modify-write therefore zeroed the toolbox "simple config" metadata. Now delegates to ManufacturerData.from_bytes, which parses the layout correctly.

M5 — public from_bytes SIZE constants disagreed with firmware (🟠)

DisplayConfig.SIZE 66→46, PowerOption.SIZE 32→30, DataBus.SIZE 28→30. The main parse path used the correct private sizes, but the public APIs rejected/misparsed real packets, and DataBus.from_bytes produced a 12-byte reserved that fed the C8 short-packet bug. Sizes and reserved slices aligned to the firmware structs.

M6 — tx_power sign confusion (🟠)

Parsed signed in the BLE path, unsigned in from_bytes, packed signed on serialize (firmware field is uint8). 0xF4 became -12 one way and serializing 244 raised struct.error. Now unsigned end-to-end.

Test plan

• uv run pytest -q → 445 passed (6 new tests: mfr round-trip, public sizes, DisplayConfig/DataBus from_bytes, short-reserved padding, tx_power unsigned round-trip)
• ruff, mypy clean

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 88.88889% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/opendisplay/protocol/config_serializer.py	85.71%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!