feat(protocol): carry HW revision + FW identity in the MSG_OK_READY ack (CAP-02)

[henols]

Firmware half of the CAP-02 pair. Must land together with henols/firestarter_app#45 — a host on this firmware and a host without it are not interchangeable.

What changed

init_programmer_framed extends the operation-setup ack past CAP-01's 2-byte buffer-size region:

[buffer_size u16 BE][effective_hw_revision u8][ver_len u8][ver bytes]

That is the whole diff — one packed emit replacing LOG_OK_ID_U16(MSG_OK_READY, DATA_BUFFER_SIZE).

Why

The host must refuse to program chips that route VPP to bus line 11 (socket pin 21 on the DIP24_2716 / DIP24_2532 pinouts) unless the attached shield is Rev 2.2 or later — those parts need the 3-position JP4 header, and driving them on an earlier board is a chip-damage path.

The revision was already computed at boot (rurp_detect_hardware_revision() in setup()), but only reachable via a separate CMD_HW_VERSION command or a FLAG_VERBOSE-gated INFO frame. Folding it into MSG_OK_READY lets the host retire its dedicated CMD_FW_VERSION pre-probe, removing one full command exchange per connect.

Notes for review

• No catalog change, no codegen regen. MSG_OK_READY is already declared params = [{ type = "bytes" }], so the wire shape was already permissive. codegen.py --check confirms the catalog is untouched.
• The revision byte is always emitted. Builds without HARDWARE_REVISION send 0xFE — the REVISION_UNKNOWN value, spelled as a literal because the symbol itself lives inside that same #ifdef. The ack's shape therefore never varies by build configuration, only by version-string length, and "no detection compiled in" reaches the host as a rejectable value rather than an absent field.
• Backward compatibility. Hosts predating CAP-02 test len(params) == 2, miss, and fall back to their 512-byte chunk floor: degraded throughput on Leonardo, never a misparse.

Size

target	flash before	flash after	delta	static RAM
uno	23954	23988	+34 B	unchanged

The 36-byte pack buffer is stack-local to init_programmer_framed, so it does not appear in static RAM — peak stack at that call site grows by ~36 B. Leonardo builds green at 90.9% flash.

Not done

• No bench validation. The wire format was verified by reading both sides and by host-side unit tests that mirror this packing — never against a real board. This is a safety change; it wants a bench run before it ships.
• This PR fires no CI. build.yml only triggers on PRs targeting main, and beta-build.yml only on push to beta. Both targets were built locally (pio run -e uno, -e leonardo).

🤖 Generated with Claude Code