feat(engine): wire the per-filament volume-map producer pipeline

- Print::update_filament_maps_to_config takes filament/volume/nozzle
  maps, backfills an empty volume map from extruder types, rebuilds
  filament_map_2, re-expands the per-filament variant arrays, and
  recomputes retract overrides keyed by resolved slots
- grouping writes its result back in every non-sequential mode;
  manual multi-nozzle grouping validates the user mapping and raises a
  translatable error on deviation; the engine's concrete volume
  assignment is deliberately not merged yet (per-filament arrays are
  already consumed by filament id, so materializing High Flow now
  would change motion before the layer-aware resolvers land)
- Print::apply treats the three map keys as engine outputs in auto
  modes (erased from the diff and adopted), compares them against used
  filaments in manual mode, and keeps the pre-expansion snapshot in
  sync with the late normalization pass so rebuilt headers reflect the
  sliced state instead of resurrecting stale values
- volume/nozzle maps and extruder_nozzle_stats join the invalidation
  group of filament_map (wipe tower + skirt/brim)
- PresetBundle composes full configs with an optional per-filament
  volume map (plate map, else defaults derived from each extruder's
  flow type); project config keeps the map sized across filament
  count changes
- PartPlate stores per-plate volume/nozzle maps; Plater injects them
  at every slice-composition site (incl. g-code reload and wipe-tower
  estimation); BackgroundSlicingProcess reads engine results back to
  the plate in auto modes
- per-filament map trust guards relaxed to size-match everywhere now
  that every producer sizes the map; single-filament explicit flow
  assignments are honored
- tests: grouping volume maps stay concrete, merge semantics of
  update_used_filament_values, single-filament override honoring

Motion g-code is byte-identical fleet-wide including Hybrid projects
(19-fixture gate + repro determinism double-slice). Header deltas:
the map keys now dump real values, and stale pre-normalization values
(e.g. enable_prime_tower on single-used-filament prints) no longer
leak into the config block.

src/slic3r/GUI/BackgroundSlicingProcess.cpp

@@ -231,7 +231,20 @@ void BackgroundSlicingProcess::process_fff()
         if (m_current_plate->get_real_filament_map_mode(preset_bundle.project_config) < FilamentMapMode::fmmManual) {
             std::vector<int> f_maps = m_fff_print->get_filament_maps();
             m_current_plate->set_filament_maps(f_maps);
+            // Orca: the volume map is not read back to the plate yet. The engine's concrete
+            // volume assignment is not written into the print config until the layer-aware
+            // g-code resolvers consume it (see the ToolOrdering write-back), so the print-side
+            // value here is just the injected plate-or-default map echoed back; persisting it
+            // would freeze the per-extruder defaults (including the transient Hybrid seed) into
+            // the plate and the saved project. Restore together with the write-back merge.
 		}
+        if (m_current_plate->get_real_filament_map_mode(preset_bundle.project_config) != FilamentMapMode::fmmNozzleManual) {
+            // The engine-resolved nozzle map is read back for every non-nozzle-manual mode so the
+            // plate config the next apply overlays matches the engine's written-back state
+            // (otherwise the full-config diff would invalidate the g-code on every apply).
+            std::vector<int> f_nozzle_maps = m_fff_print->get_filament_nozzle_maps();
+            m_current_plate->set_filament_nozzle_maps(f_nozzle_maps);
+        }
 		wxCommandEvent evt(m_event_slicing_completed_id);
 		// Post the Slicing Finished message for the G-code viewer to update.
 		// Passing the timestamp