chara_fur: 91.76% -> 91.78%

[SirEnkido]

What changed

Two matching improvements in src/chara_fur.cpp:

• CChara::TimeMogFur: reordered the tileIndex address-computation terms (tile-row, tile-col, in-tile-row, in-tile-col) so MWCC's loop-invariant code motion hoists the same subexpressions as the target
• CChara::InitFurTexBuffer: hoisted idx/row to function scope with split declaration/init to match the target's declaration-order saved-register allocation, which made the scoped #pragma opt_propagation off unnecessary — removed it

Evidence

• main/chara_fur fuzzy (report.json): 91.764% → 91.778%
• DOL verified byte-identical (build/GCCP01/ok passes)
• Rebuilt in a clean worktree and independently re-verified before submission

Notes for review

• Small gain on a hard unit; both changes also remove a pragma crutch, which we believe is net-positive for future matching here.
• The pre-existing reinterpret_cast<unsigned char*> byte addressing on m_texels (GX texture-tile addressing) predates this branch and is untouched.

---

Produced by Claude agents following AGENTS.md / WORK_SPLIT.md, operated by @SirEnkido.

🤖 Generated with Claude Code