RGBGFX reverse mode checks for too-low palette IDs, not just too-high

[Rangi42]

I'm trying to fix the sanity checks for invalid input into RGBGFX reverse mode. Specifically, we should ensure that the tilemap's tile IDs and the attribute map's palette IDs are valid, even with nonzero base tile/palette IDs specified.

However, I'm having trouble getting that to cooperate with the randtilegen/rgbgfx_test randomized test cases. Sometimes they fail, as in https://github.com/Rangi42/rgbds/actions/runs/29282085627/job/86927774014:

Testing: ./rgbgfx_test seed8.bin -b 217 -N 57,256
error: Color mismatch at (8, 56): (90,198,148,255) became (66,66,214,255) after round-tripping
error: Color mismatch at (9, 56): (231,173,0,255) became (16,173,107,255) after round-tripping
...
error: Color mismatch at (61, 63): (132,148,247,255) became (189,165,57,255) after round-tripping
error: Color mismatch at (63, 63): (189,165,57,255) became (0,24,140,255) after round-tripping
Test failed with 342 errors
Test ./rgbgfx_test seed8.bin -b 217 -N 57,256 failed! (RC=1)

Edit: Turns out it was a mistake in 21eaa2d, which 8b951bc has fixed. I confirmed this by locally editing test/gfx/test.sh to iterate over every possible $ofs value ("All 30723 tests passed!").