feat: two-colour AMS jobs, remote filament load; resume confirms RUNNING

[voska]

Summary

Two-colour printing without anyone at the printer, plus a remote colour change at a pause:

• Two-filament jobs with an automatic AMS change at a layer height.
  • slice --filament P1 --color C1 --filament P2 --color C2 --filament-change-z Z writes the change the way Bambu Studio's layer slider stores it (--load-custom-gcodes, a ToolChange in MultiAsSingle mode), so the G-code carries M620 S1A / T1 / M621 S1A at that layer and the AMS swaps by itself.
  • slice reads the change back from the G-code and exits 11 without touching the output dir if Studio didn't place it. Studio drops a ToolChange silently when the mode doesn't fit.
  • preflight / print send take one --slot per filament. Every per-filament gate runs against that filament's own slot, and ams_mapping becomes e.g. [3, 0].
• filament load --slot S --confirm swaps the toolhead to another AMS slot over LAN (print.ams_change_filament). It works only while PAUSE or IDLE/FINISH/FAILED (an allowlist). When paused, the toolhead must hold an AMS filament and the slot must hold the same material. It exits 0 only once the printer reports S in the toolhead with the change finished.
• Fix: print resume used to exit 0 three seconds after the ack even if the printer was still paused. It now waits for RUNNING (--wait, default 30s, must be positive; otherwise exit 14). It refuses during a filament change, and on X1-series printers when the AMS reports an empty toolhead. That's the only place tray_now 255 is verified to mean empty.

Contract

• Additive only:
  • status.ams_status
  • slice filament_changes / prime_tower / filament_presets, and per-filament model_g / waste_g / nozzle_temp / nozzle_temp_range
  • preflight slots / tray_ids, and filament on gate rows of multi-filament jobs
  • job filament_changes
• Single-filament jobs are unchanged. Against skunkworks, print send --dry-run --slot A4 on the same file gives a byte-identical MQTT payload and FTPS plan with v0.2.0 and with this branch, and the same gate names and statuses. Only the single_filament detail text differs.
• Exit codes: print resume can now exit 14 (not RUNNING in time) or 9 (mid-change / empty toolhead) where it used to exit 0.

Verification

• make ci passes (fmt-check, golangci-lint 0 issues, vet, go test -race, build).
• New tests run on recorded fixtures: a real G-code pause report, and trimmed real 1- and 2-filament Bambu Studio 02.08 slices with no serials. A test run against main shows print resume --confirm exiting 0 with "state_after": "PAUSE".
• Real slice: the 27-layer button sliced white (Generic PLA) through Z 4.0 and black (PLA Matte) above it with Bambu Studio 02.08.02.61. Read back with unzip | awk, the only in-print tool change is M620 S1A / T1 / M621 S1A at layer 21 (Z 4.2), plus the end T255. There's no pause. Filament 1 16.98 g (16.02 + 0.97 waste), filament 2 3.06 g (2.89 + 0.18), 44 min 43 s.
• Live and read-only: against skunkworks, print send --dry-run --slot A4 --slot A1 gives PASS, with one WARN (A4 has no RFID, so its remaining filament is unknown) and ams_mapping [3, 0]. With --slot A4 alone it exits 9; with the same slot twice it exits 2. Nothing was uploaded or published.
• Not verified on hardware: a real two-colour print (ams_mapping [3, 0]), and the Go filament load itself. The identical payload was verified on this printer from bambu-op.