fix(gl): derive the GL index type from sizeof(ofIndexType) in indexed draws

[smeyfroi]

Problem

On current macOS nightlies (verified against of_v20260730, macOS 26 / Apple silicon), every indexed VBO draw silently reads the index buffer with the wrong GL type, and instanced indexed draws (ofVbo::drawElementsInstanced) render garbage triangles or nothing at all, varying non-deterministically between launches and even between relinks of the same code.

Cause

ofIndexType is typedef TESSindex ofIndexType (ofConstants.h), and the tess2 header shipped since apothecary #552 picks its width with #if defined(TARGET_OS_IPHONE) || …. Apple's TargetConditionals.h defines every TARGET_OS_* macro on every Apple platform — as 1 or 0 — so on macOS TARGET_OS_IPHONE is defined as 0, the defined() test is true, and TESSindex/ofIndexType becomes unsigned short. (Apothecary #558 later noticed Apple builds landing in the 16-bit branch, but guarded only the __arm__-family defines, which aren't defined on Apple arm64 — the actual trigger survived.)

Meanwhile the desktop branches of drawElements / drawElementsInstanced in both renderers hardcode GL_UNSIGNED_INT. ofVbo uploads sizeof(ofIndexType) × count bytes, so the draw reads 2× past the uploaded index buffer: the first indices are misparsed short-pairs (out of range of the mesh), the rest are whatever follows in GPU memory — out-of-range vertex fetches that render as giant garbage triangles or degenerate/invisible ones depending on heap layout. No GL error is raised.

Fix

Derive the GL index type from the compiled index width at the five desktop call sites:

sizeof(ofIndexType) == 2 ? GL_UNSIGNED_SHORT : GL_UNSIGNED_INT

The ternary constant-folds, so there is no runtime cost, and the sites remain correct whatever width tess2 ends up giving ofIndexType. The GLES branches are untouched: every GLES platform lands in tess2's 16-bit branch, so their hardcoded GL_UNSIGNED_SHORT already matches.

A companion apothecary PR fixes the root-cause condition in tess2.patch itself: openframeworks/apothecary#561. Both fixes stand alone; this one keeps the renderers honest under either header, and the apothecary one also restores the full 32-bit index range for large meshes on macOS.

Verification

• Reproduced on macOS 26 / Apple silicon with an instanced line renderer: 16-bit uploads (12-byte index buffer for a 6-index quad, confirmed via GL_BUFFER_SIZE) drawn with GL_UNSIGNED_INT produce garbage or missing geometry, flipping between relinks.
• With this patch, disassembly of both Release build systems (Xcode and makefile) shows the draw sites passing GL_UNSIGNED_SHORT (0x1403), matching the uploads; the renderer output is correct and stable across relinks.
• The pre-CRITICAL FIX: ios touch positions are in the wrong place when in landscape #552 tree (of_v20251123, 32-bit ofIndexType) is unaffected by the patch: the ternary folds to GL_UNSIGNED_INT there.