feat(gust): cross-arch matrix — one wasm lowers to ARM and RISC-V with the seam intact

[avrabe]

Moves the biggest item off the Research Day cut's WON'T list.

What changed

Until this morning, no thin-seam driver crossed to RISC-V: synth's RV32 backend declined every function containing an imported call. That was filed as synth#871 and fixed the same day in synth 0.52.0 (auipc/jalr + R_RISCV_CALL_PLT into .rela.text).

Re-measured on our actual drivers:

driver	ARM cortex-m3	RISC-V rv32imc
wdg-thin	6 T, 2 U, 638 B	5 T, 2 U — 1 of 6 skipped
gpio-thin	5 T, 2 U, 502 B	5 T, 2 U — complete
i2c-thin	7 T, 2 U, 954 B	6 T, 2 U — 1 of 7 skipped
timer-thin	5 T, 2 U, 204 B	5 T, 2 U — complete
adc-thin	8 T, 2 U, 736 B	8 T, 2 U — complete
dac-thin	8 T, 2 U, 642 B	8 T, 2 U — complete
pwm-thin	6 T, 1 U, 694 B	6 T, 1 U — complete
uart-thin	4 T, 3 U, 254 B	4 T, 3 U — complete

Six of eight lower completely on both, same function count, same seam width. Deterministic.

The gate is the undefined-symbol count, not size

Under 0.49, wdg-thin lowered to a 468 B RV32 object with 2 defined functions and zero undefined symbols — smaller than the ARM object, and hollow: every gust:hal call had been silently dropped. So build-cross-arch.sh fails if either object has no undefined seam symbols. A smaller object is exactly what the failure looks like.

Honest boundary (in the doc, not just here)

• RISC-V objects are compiled and seam-correct, not executed — and could not meaningfully be: these carry STM32 register maps, so a faithful RV32 silicon result needs an RV32 target model that does not exist yet.
• ARM remains the only executed leg, on two dies.
• gust_mix already has real ESP32-C3 silicon numbers (1.839x). That is execution evidence; this table is lowering evidence. Different claims, kept apart.
• Two functions still decline, reported as synth#882.

Also measured

0.52.0 shrank every ARM object, no regressions — −122 B across the suite, including gpio-thin 534 → 502 B (relevant to synth#846). This PR does not re-pin the toolchain; the committed .o files stay on 0.49 pending a silicon re-run.

🤖 Generated with Claude Code

https://claude.ai/code/session_011QG86sovTbfnPNY9SfhSmo

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!