[DarkBiscuits]

A custom mill tool-change template containing G53 G0 Z0 posts and passes read-back, but mill audit rejects it as unsupported G53.

Accept G53 in mill audit as a non-modal machine-coordinate move, excluded from part-frame motions. Preserve the preceding part position, motion mode and plane so a machine retract does not corrupt subsequent modal lines or arcs. Validate the block's words before skipping its machine-coordinate motion. This follows the existing read-back/five-axis treatment and does not model a machine-to-work transform or verify the physical retract path.

Regression coverage includes a posted custom tool-change template with IR round-trip, modal arcs following retracts (including reordered words and plane words), malformed-word rejection, and randomized machine coordinates followed by a partially specified modal feed move. The three positive regression tests failed with unsupported G53 before the fix.

Validation on Linux (Rust 1.95):

• All four focused mill_g53 tests passed, including the property test.
• Complete unmodified scripts/check.sh passed: formatting, workspace Clippy with warnings denied, workspace tests, wasm library checks, and licenses.
• Re-ran the original synthetic ring-pocket CLI case and audited/read back both the built-in Mach3 output and custom G53 output successfully, with identical part-motion counts.

AI disclosure: implementation, regression tests, review, and validation were performed with AI assistance using CAMpanion's own sources and synthetic inputs, following the clean-room contribution rules. No physical controller testing was performed.

This branch starts from upstream main and is independent of #152.

Fixes #145.

[mountainarcwelding]

Thanks for these, really appreciate the quick turnaround and the tests. I've kicked off CI and will review once it's green. One small ask: could you change "Fixes #145" to "Refs #145"? I keep issues open until a fix has been run on a real machine, so I'd rather they not auto-close on merge.