stb_image speedup notes

gistfile1.txt

png (test.png)
==============

initial:                     30 runs in 804.013 ms -> 26.800 ms/run
improve filter:              30 runs in 684.728 ms -> 22.824 ms/run (-14.8% vs. original)
zout in local var:           30 runs in 514.701 ms -> 17.157 ms/run (-36.0% vs. original)
single-byte runs:            30 runs in 483.597 ms -> 16.120 ms/run (-39.9% vs. original)
extract zhuff_dec slow path: 30 runs in 469.272 ms -> 15.642 ms/run (-41.6% vs. original)
improve 'fast' table:        30 runs in 451.185 ms -> 15.039 ms/run (-43.9% vs. original)
guess decoded size:          30 runs in 386.372 ms -> 12.879 ms/run (-51.9% vs. original)