Fix RGB->BGR channel flipping. First 4 pixels were being skipped.

[ryanrsrs]

Caused first 4 pixels of every line to have bad colors. Not noticable unless the image has bright colors along the left edge.

[jeffrbig2]

Gad, the scalar paths of the channel flipping weren't being tested. Sorry, will update. What platform were you on that the SIMD path wasn't selected, btw...?

[ryanrsrs]

"armv7l" i.e. 32-bit ARM Raspberry Pi Zero 2 W.

Code was compiled with CFLAGS including: -march=native -Os -DSTBIR_USE_FMA -mfpu=neon-vfpv4 -mfp16-format=ieee

Here is my makefile: https:/ryanrsrs/console-jpeg/blob/main/Makefile

I tried with and without -DSTBIR_NO_SIMD, but I think this scalar path was always being used. Should it be NEON?

[jeffrbig2]

Oh right, that path has no simd path now. Let me do your fix and add one...