Document the full esp32-doom performance story: the bottleneck (per-event SPI
C->Python ctypes crossings, not QEMU compute / libqemu -O level / transport /
core version), the dead ends that gave zero gain and why (-O2 rebuild, async
_emit, blind CS-flush suppression), the two fixes (batch SPI data, gate CS
crossings) for 0.04 -> ~1 FPS (~26-37x), the build/test playbook, key files and
remaining headroom (the DC pin).
David Montero Crespo