Decode negative HTTP body integers consistently

[barryw]

Negative one-, two- and three-byte HTTP body integers decode as positive when plain char is unsigned. Use int8_t for the sign byte and multiplication instead of shifting a negative signed value.

Introduced after 3.14 with the HTTP target. Tests cover zero, maximum positive, minimum negative and -1 at all four wire widths. Tested independently against test-merge at 8596576f, in the upstream pipeline’s my_docker_image (host G++ 13.3.0):

make -B -C target/pc/linux/test_http_target test-integer-widths

Same tests with unsigned char, changing only the firmware fix:

Before: Integer widths: 99 checks, 6 failures; exit 2
After:  Integer widths: 99 checks, 0 failures; exit 0

Red/green evidence; final CI build passed all five firmware targets and the application-space gate. Hardware E2E was not run.

[chrisgleissner]

@barryw Thanks for spotting this.

Could you please analyze which actual user facing features are degraded by this bug so we can triage it for 3.15?

[barryw]

@chrisgleissner I think this is a 3.16 bug. I just don't like leaving broken code lying around. We can leave it until after 3.15 ships.