box/aabb2d: fix wrong face selected below AABB min bound in sphere/circle test

[94xhn]

Problem

glm_aabb_sphere() (include/cglm/box.h) and glm_aabb2d_circle() (include/cglm/aabb2d.h) compute the closest point on an AABB to a sphere/circle center per axis using:

a = (s[i] < box[0][i]) + (s[i] > box[1][i]);
...
box[!(a - 1)][i]

a is 1 for either out-of-range direction (below box[0] or above box[1]), and !(a - 1) collapses to index 1 (the max face) whenever a == 1, regardless of which side the point is actually on. So when the query point is below the box's min bound on an axis, the code clamps to the max face instead of the min face, producing an incorrect (too-large) distance-squared contribution and false-negative intersection results.

Fix

Use the already-computed "above max" boolean directly as the face index instead of the !(a - 1) derivation:

box[s[i] > box[1][i]][i]

This selects box[0] (min face) when the point is below the min bound and box[1] (max face) when above the max bound. The in-range case is unaffected — it's already zeroed out by the existing (a != 0) multiplier, so the index value there doesn't matter.

Same one-line-per-axis change applied identically to both glm_aabb_sphere (3 axes) and glm_aabb2d_circle (2 axes), since they share the exact same root cause.

Verification

• Manual cases: sphere/circle center below box[0] on one axis, below on two axes simultaneously (3D), etc. — all mismatched (false negative) before the fix, correct after.
• Randomized fuzz test, 200,000 cases with random box bounds / sphere center / radius, checked against a hand-written point-to-AABB squared-distance reference: 3418/200000 (1.7%) false negatives before the fix, 0/200000 after.
• Full cglm.h still compiles clean with -Wall -Wextra.

Diff is minimal: 2 files, 5 insertions / 5 deletions, no behavior change to the in-range or above-max cases.

[recp]

@94xhn the PR is merged, thanks for your contributions 🚀