optimize a lot

[amaanq]

we love perf gains

commit	size (B)	syscalls	instructions	p5 wall
main	28,112	116	115,357	137.1 µs
lib: loop over colors array in write_dots	27,424	116	115,329	137.6 µs
cpu: deduplicate /proc/cpuinfo read	27,432	116	115,284	135.5 µs
lib: pack the NixOS logo into a flat byte stream	22,912	116	120,517	136.0 µs
lib: switch write_row to dyn FnMut	20,032	116	121,310	135.8 µs
lib: shrink output pipeline and fix runtime ABIs	14,656	116	120,092	135.8 µs
asm: copy and fill memory a word at a time	14,736	116	60,740	132.1 µs
cpu: probe the ends of the CPU list for the max clock	14,960	28	44,482	87.4 µs
cpu: read only the first /proc/cpuinfo record	14,960	28	43,842	82.8 µs
total	−46.8%	−75.9%	−62.0%	−39.6%

The first five commits are size work and their wall times are all inside the noise floor of each other. The speed improvements comes from the last three commits.

[NotAShelf]

We sure do, wanna cut a new release now?

[amaanq]

sure