fix(fast3d): don't clamp HD textures to an undefined tile region

[bassdr]

Summary

With an HD texture pack, characters whose CI eye/face textures are loaded via segment with an unset tile size render those textures as a small repeated grid. Most visible on adult Zelda / Sheik in the Temple of Time. Vanilla textures are unaffected; it only appears with HD replacements enabled.

Root cause

Regression from #1118. That change made HD replacement textures always clamp their sampled width/height to the rendered tile region (tex_width2/tex_height2) via the isHd flag in GfxSpTri1.

These eye textures use a tile with no size set (lrs == uls == 0), so tex_width2 = (lrs - uls + 4) / 4 = 1. The isHd clause then clamps tex_width from its real 32 down to 1. Since tex_width is also the UV normalizer (u / tex_width), the vertex UVs (authored 0..32) now span 0..32, and with wrap addressing (cms == 0) the upscaled texture tiles ~32x across the primitive.

Instrumentation at the clamp, one HD CI8 draw (broken eye i=1 vs. a face tile i=0 with a real region):

i=1  hbs=32 vps=8  tw=32->1  tw2=1   uls=0 lrs=0    cms=0 cmt=0  siz=CI8   <- clamped to 1
i=0  hbs=64 vps=16 tw=192    tw2=192 uls=0 lrs=764  cms=2 cmt=2  siz=CI8   <- no-op (correct)

Before #1118 these textures rendered fine: they are wrap-mode and not pyramid-like, so the tile-region clamp never applied.

Fix

Only force the HD tile-region clamp when the tile actually defines a region (lrs > uls / lrt > ult). The pyramid-like and explicit G_TX_CLAMP paths are unchanged, and HD textures that genuinely define a smaller sub-region (the case #1118 targeted) still clamp as before.

Scope and testing

Affects any HD pack drawing actors that segment-swap a CI eye/face texture with an unset tile size (adult Zelda, Sheik, likely others). Tested with an HD pack on the Temple of Time Sheik-reveal cutscene: the eye renders correctly after the change, with no regressions seen on other textures (HUD, world, faces with real tile regions).