Try AMY CPU overload detection + failsafe (amy#826)

[bwhitman]

Pins the amy submodule to the head of shorepine/amy#826 (CPU overload detection + failsafe, issue shorepine/amy#791) so this PR's firmware and webapp preview builds carry it for hardware testing.

Also adds tulip.amy_render_load() → float in modtulip.c: AMY's smoothed fraction of real time spent rendering (0..1, ~1.0 = overloaded), backed by the new amy_get_render_load(). Poll it while a heavy sketch runs to watch the load climb.

What to try on AMYboard with this build:

• Run dpwe's Eno sketch (or anything heavy). As it stacks up voices, tulip.amy_render_load() should climb toward 1.0.
• At sustained ≥0.98 for 250ms, AMY resets itself and plays four descending doots instead of wedging — and USB CDC / MIDI should stay alive throughout (the fill-buffer task now yields when the i2s write stops blocking).

Follow-up (not in this PR): implement config.amy_external_overload_hook in tulip to stop the running sketch and pop up "this sketch took too many resources" in AMYboard Online.

🤖 Generated with Claude Code

[github-actions]

🔌 AMYboard PR preview

Editor + flasher: https://amyboard-pr-1105.vercel.app/editor/

This preview bundles this PR's firmware — its flasher only flashes this build (not the release). Rebuilt on every push; removed when the PR closes.

The hardware CI has been kicked off and should return within a few minutes, stand by!

[github-actions]

🌷 Tulip Web PR preview

Tulip Web: https://tulip-pr-1105.vercel.app/run/

Flash a Tulip to this build: on-device run tulip.upgrade(pr=1105).

Rebuilt on every push; removed when the PR closes.

[bwhitman]

Added the full overload chain in b09a5c6 — event-driven end to end, no polling anywhere:

amy_fb_task trips failsafe (synth already reset, doots played)
  → tulip_amy_overload_hook: mp_sched_schedule(callback, load%)   [amy_connector.c]
    → amyboard._on_amy_overload on the MP task:
        stop_sketch()                                              [so it can't re-wedge]
        F0 00 03 45 'L' <b64 load%> F7 out the USB-MIDI gadget     [best-effort]
          → AMYboard Online's persistent sysex listener             [spss.js]
            → "This sketch needed more CPU than the AMYboard has…" modal

New Python API: tulip.amy_overload_callback(cb) (pass None to clear). The load crosses the task boundary as a percent small-int, so no cross-task heap allocation. Headless boards just get the sketch stopped — the doots were the notification.

🤖 Generated with Claude Code

[github-actions]

🎛️ HW CI (physical bench)

AMYboard (USB-MIDI + AMY zP → audio; built-in tones + AMYboard World sketches acid/house/woodpiano over the SysEx control API): ✅ PASS — flashed this PR’s firmware; all checks matched the references.

Tulip (TULIP4_R11; serial-REPL audio + WiFi screenshot): ✅ PASS — flashed this PR’s firmware; all checks matched the references.

⬇️ Artifacts: recordings · screenshot · serial logs · run logs

_{Self-hosted bench. Audio spectral-compared to ref/hwci_basic.wav, the AMYboard World sketch refs (ref/{acid_generator,house_generator,woodpiano}.wav) + ref/tulip_basic.wav; Tulip screenshot pixel-compared to ref/tulip_screenshot.png. Both analog outs share one capture card, so the tests run sequentially.}