fix: visualizer renders G18 radius-format arcs correctly

[misiekhardcore]

Summary

Fixes the visualizer rendering G18 (ZX) plane arcs as straight lines. Reported with clamex_p14_slot.nc - a three-pass semicircular groove carved via G18 + G2 X±w Z0 R50.376 that rendered as 3 straight lines.

Two bugs fixed:

1. Radius-format (R-word) arcs were ignored. The extractor only read I/J/K center offsets; G2 X21.881 Z0.000 R50.376 had no offsets, so the computed radius was 0 and the arc degenerated into a straight line. Added resolveArcOffsetsFromRadius, which reconstructs the arc center from chord geometry and picks the candidate whose sweep matches the G2/G3 direction and the R sign (positive R = short arc <= 180°, negative R = long arc > 180°). A shared computeArcSweep helper deduplicates the sweep normalization with generateArcPoints. Fallbacks: I/J/K offsets take precedence when non-zero; impossible geometry (|R| < chord/2, zero radius, start == end) still degrades to a straight line.

2. G18 arc direction was inverted. The XZ plane config used a left-handed in-plane axis order (X, Z). The standard convention (CW/CCW as viewed from +Y, per LinuxCNC/NIST) requires the cyclic order (Z, X) so the interpolation angle increases CCW about the positive normal axis uniformly in all three planes. Swapped the config to (z, x) with K/I offsets; G17 and G19 untouched.

Verification

• The reported file pattern now extracts 3 ARC_CW segments dipping to Z = −5.0 / −10.0 / −14.0 mm - verified both via a synthetic regression test mirroring clamex_p14_slot.nc and by running the actual file through the fixed extractor.
• New tests: G18 direction (G2 dips below the chord, G3 bulges above), G17/G19 regression guards, R-word arcs in all three planes, positive/negative R (short/long arcs) for both G2 and G3, exact-semicircle negative R, G91 incremental, impossible-radius fallback, I/J/K-over-R precedence, and the clamex 3-pass groove test with sampling-robust depth bounds.
• Validation: tsc --noEmit clean; extractor suite 91/91; full unit suite 85 suites / 1417 tests; lint clean on touched files; e2e suite 96 passing.

AC checklist

• R-word (radius-format) G2/G3 arcs render as interpolated arc segments in G17, G18, G19
• Negative R renders the long arc; positive R renders the short arc
• G18 arc direction follows the standard convention (G2 dips below the chord as viewed from +Y)
• The clamex 3-pass pattern renders as 3 semicircular passes at Z = −5 / −10 / −14
• Existing behavior preserved: I/J/K offset arcs unchanged; no-offset/no-R and impossible-R cases still degrade to a straight line
• CHANGELOG updated; lint, typecheck, unit suite green

Commits

• 657503b fix: correct inverted G18 (XZ) arc direction in visualizer path extractor
• 1c61b9a fix: render radius-format (R-word) G2/G3 arcs as arcs in visualizer
• 8e9562b fix: render negative-R semicircles and deduplicate arc sweep computation
• f56c8b3 test: harden clamex groove-depth assertions against arc sampling parity