feat(export): slicer-ready 3MF (plate layout, per-part colour/material, slicer sidecars, Bambu .gcode.3mf)

[w1ne]

Slicer-ready 3MF export: one file that opens in a desktop slicer with every part as a named object, its colour and material kept, the parts laid out on the plate, and optionally a filament slot per colour.

What the 3MF wrote before

• One <object> per part, with names kept.
• One <basematerials id="1"> entry per part (name = part name, displaycolor from .color()/.material()). Role tokens such as 'tool' fell back to #CCCCCCFF.
• The colour was set on every triangle (pid="1" p1="i"). Objects had no pid/pindex.
• Build items had no transform, so parts stayed at their raw world-frame positions: they were not on Z=0 and could overlap.
• No slicer metadata.

What it writes now

• Core materials: objects carry pid/pindex at object level. There is one <base> per distinct (name, colour), and the name is the part's engineering material (arm.part(..., { material: 'pla' })) or else the part name. Role tokens resolve through the shared palette. To support this, materialName is added to the assemblyPart record metadata, SceneBackendPart and WorldFramePart.
• options.arrange:
  • none (default): same positions as before.
  • plate: one build item per part, dropped to Z=0 and shelf-packed on the printer bed with 5 mm gaps and no overlap.
  • assembled: one multi-part (components) object for multi-colour inlays, dropped to Z=0 and centred on the bed.
• options.orient (plate only): turns each part so the largest flat face it can rest on faces down. A face counts only if no vertex lies beyond it, so a cup's inner floor never wins.
• options.slicer:
  • generic (default): no sidecar.
  • bambu / orca (same format): Metadata/model_settings.config with the name and filament slot (extruder) for each object and each part, all on plate 1.
  • prusa: Metadata/Slic3r_PE_model.config with one volume per part by triangle range.
  • Slot N is the Nth distinct (material, colour) pair.
  • On purpose, there is no project_settings.config: a partial one crashes the loader in a real slicer (confirmed locally), and a full one would override the user's presets.
• send_to_printer with bambu-lan: the upload is now a .gcode.3mf. The G-code goes in as Metadata/plate_1.gcode (plus an .md5), which is the path the existing MQTT project_file command already names. It is packed inside the model 3MF given as model_3mf_path (new; also kernelcad print send --model-3mf), or else inside a minimal core-3MF shell. The file name always ends in .3mf.
• Attribution: the model's core Application metadata is now attributionGenerator() from src/shared/links/attribution.ts (kernelCAD <v> (https://kernelcad.com)), plus Description = Made with kernelCAD (https://kernelcad.com). A test reads both back.
• Tool prose, kernelcad-mcp and kernelcad-print skills, the public-contract fixture, and a new slicer-ready-3mf cookbook snippet.

Tests

• tests/unit/kernel/backends/occt/export3mfSlicer.test.ts (12): parses the XML with a namespace-aware parser and checks:
  • core namespace, the basematerials resource, and valid object-level pid/pindex and build refs
  • colours round-trip (hex and role token)
  • plate: parts on Z=0, inside the bed, no XY overlap, also when parts overflow one row
  • orient lays a slab flat and leaves a part that is already flat unchanged
  • assembled keeps relative offsets
  • the bambu and prusa sidecars match the model, and generic writes none
  • bad arrange/slicer values are rejected
  • the material name reaches the file end-to-end via runAndExport
• tests/unit/kernel/print/bambuPrintFile.test.ts (4) and 2 new send_to_printer cases.
• 12 of the new tests fail on develop. The other 6 are regression guards and pass on both.
• The existing 3MF tests and the eval harness pass.

Checked in a real slicer (only partly)

With a locally installed slicer based on the same 3MF loader, all 6 combinations (plate/assembled × generic/bambu/prusa) load as manifold objects. In assembled mode they load as one object with N parts. The bambu sidecar round-trips per-part names, per-part filament slots (1/2/3) and the plate positions. Not tested: the actual Bambu Studio, OrcaSlicer and PrusaSlicer binaries, and a real Bambu printer.