v0.6.1: engine changes no longer stall the audio (#36)

[Worldwave]

Fixes #36.

The bug

Changing engine could stall the audio for up to ~4 ms inside a 2 ms block. Plaits' engines share one block of RAM, so each engine's Reset() clears its buffers on every change. GCC sends those fills to picolibc's memset, which on this toolchain is a byte loop at ~6.6 cycles per byte:

• Particle: 16 KB diffuser, ~1.7 ms.
• Particle, entered with MORPH below centre: twice that. The M3f diffuser-skip override clears again when diffusion resumes.
• String: 10 KB of delay lines, ~1.1 ms.

The fix

• firmware/src/sp1_zero.cc: plaits::sp1_zero fills a word at a time, with byte head and tail for unaligned buffers. It's built with -fno-tree-loop-distribute-patterns so GCC can't turn it back into a memset call.
• Two one-passage overrides: FxEngine::Clear() (Particle's diffuser, the ensemble used by String machine and Chords) and DelayLine::Reset() (String). 14 call sites, the same byte counts as before.
• Deliberately scoped to Plaits. libc's memset is not replaced: it also zeroes .bss before main(), where a fault is a boot loop on a device with no reset pin.
• tools/ci/check_image.py: fails an image where sp1_zero calls anything, or where Particle's or String's reset still calls memset.
• v0.6.1: firmware/VERSION, README, issue-form placeholders.

Verification

• Both configs: 0 warnings. Plaits build 424 420 B, links at 0x20000, check_image.py all pass.
• Host: all 8 suites pass. Renders are bit-identical to main across 30 engine changes with MORPH crossing centre, at four arena alignments. A guard-byte test of sp1_zero covers 296 alignment/length cases.
• Hardware, 2026-10-05: the same code, built before the version bump and labelled v0.6.0.

	v0.6.0	this branch
Particle, first call after a change	1984–4000 µs	457–625 µs
String, first call after a change	1559 µs	509–511 µs
Worst block, Particle/String round trips	up to 248 %	80.9 %
fail / rst	0 (6.6 ms blocks failed in development)	0

The per-block output clears, which go through the same byte loop, are #37.

🤖 Generated with Claude Code