sched_setaffinity binds one thread, so the fallback that undid a GameMode pin
only freed the thread that called Start(); the worker threads the fix exists to
protect stayed on the daemon's cores. Walk /proc/self/task instead, and treat a
readdir failure partway through as a failed restore rather than a finished one.
Also stop chrome_trace_writer emitting a valueless "ts" on to_chars failure, or
a bare inf/nan, either of which costs the whole trace instead of one field.
Release the GameMode request when the daemon narrows the CPU mask; the pin
concentrates Roblox's worker pool and is re-applied every few seconds, so
ending the request is the only thing that lifts it. `gamemode: on` keeps it.
Emit texture uploads on worker threads instead of serially inside
vkQueueSubmit, in two passes so mip 0 lands before the levels scaled from it.
Decode ETC2 through a persistent pool and lower the parallel floor to one
128x128 image, so small batches stop running single-threaded.
Document all three in docs/PERFORMANCE.md.
vii