fix(h3d-carve-14): restore dropped tw arg in tremoloOffsetWorldX call; distinguish mStr/mGlow fakes

Creed F1 (MED): slideRibbonUpdatePositions called tremoloOffsetWorldX(n, Tk) — the
move dropped the tw (ribbon width) argument. Helper signature is (n, chartTime, trailW);
with trailW=undefined every ribbon vertex position was NaN for notes with n.tr truthy
(tremolo slide sustains). Restored verbatim: tremoloOffsetWorldX(n, Tk, tw).

Creed F2 (LOW): _buildDrawNote gave getMStr and getMGlow identical fakeMat objects, so
swapping the wiring left the suite green. Now each string index gets its own named object
(mStr[i]/mGlow[i]); added behavioral assertion that gem core.material === mStrMats[s].

Tests: 26/26 note-renderer pass (+2 new kills: tests 25 and 26). Suite 1379/2 (same
2 pre-existing failures as 59bccef baseline).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014uZ169yfoFYArXz962g7KW

plugins/highway_3d/src/note-renderer.js

@@ -112,7 +112,7 @@ export function createNoteRenderer({
             const zk = dZ(Tk - now);
             const xc = strandBaseX
                 + dirMul * slideOffsetWorldX(n, Tk, slideSt)
-                + tremoloOffsetWorldX(n, Tk);
+                + tremoloOffsetWorldX(n, Tk, tw); // h3d-carve-14 creed-fix: restore dropped tw arg
             const yc = y + techniqueYOffsetWorld(n, Tk);
             pa[v++] = xc - tw * 0.5; pa[v++] = yc - th * 0.5; pa[v++] = zk;
             pa[v++] = xc + tw * 0.5; pa[v++] = yc - th * 0.5; pa[v++] = zk;