fix: regenerate tree supports on move and grow branch bases near the bed

[zackaree-shen]

Note: Supersedes #782. The head branch has been migrated to this fork per the branch cleanup policy (personal dev branches are no longer allowed on Snapmaker/OrcaSlicer).

Description

Two independent tree-support fixes:

1. Tree supports were not regenerated after moving / centering an object

• Tree support branches are routed against the machine border (bed bounds) anchored to the instance shift, but PrintObject::set_instances() only invalidated the Print-level steps (psSkirtBrim, psGCodeExport) on a pure XY translation, which never cascade to the PrintObject-level posSupportMaterial step.
• Result: after moving or centering an object, slicing kept reusing the supports computed for the old position, until some support option was toggled back and forth.
• Fix: for tree-support objects, also invalidate posSupportMaterial on instance changes. Regular supports are computed in object coordinates and stay valid on translation, so the cheap skirt/gcode-only invalidation is kept for them (no performance regression when dragging objects).

2. Tree support branches spill out of grooves/cavities (ported from BambuStudio d61ebefa2, STUDIO-18228/18439, OrcaSlicer#10542/OrcaSlicer#10775)

• Grow branch radius by half an extrusion width per layer for nodes in the first ~2 mm above the bed (when wall count is auto or dual-wall), so tree bases fill cavities/grooves instead of spilling out of them.
• DO_NOT_MOVER_UNDER_MM becomes a class constant of 2.0 mm (was a local is_slim ? 0 : 5).
• drop_nodes(): add bottom_expand_enabled gate and apply the radius growth when dropping nodes near the bed.

Screenshots/Recordings/Graphs

To be attached during manual validation (groove test model): before this PR the groove model produced 559 fine support markers spilling outside the groove (vs BambuStudio 37 stout markers inside).

Tests

• Full Release build via VS2022 (orca-build) - compiles clean, app runs
• Manual: groove model - slice, then center the object, then re-slice: supports must be regenerated (G-code support markers change)
• Manual: branch bases near the bed are thicker and stay inside the groove
• Note: automated slice tests are not runnable in this fork (pre-existing test-infra limitation, verified against a stash baseline)