perf: avoid copying rvalue meshes when adding model volumes

[VailElla]

Summary

• avoid deep-copying TriangleMesh when ModelObject::add_volume receives an rvalue
• preserve convex-hull initialization after the mesh is moved into the volume
• add a regression test that verifies the vertex and index buffers are transferred

Validation

• cmake --build build/arm64 --target libslic3r --parallel 8
• focused fff_print_tests regression: ModelObject moves rvalue mesh storage into a volume
• git diff --check

The full fff_print_tests target is currently blocked by pre-existing test sources that use an older GCodeWriter API (set_extruder, lift, and retract_lift); those failures are unrelated to this change.

[Haidiye00]

Hi, under what bug is this? Does it need to be fixed this way? Can you tell me what this bug is?

[VailElla]

Hi, thank you for taking a look.

This PR is not intended to fix a user-visible functional bug. It addresses an unnecessary deep copy on an existing rvalue code path.

ModelObject::add_volume(TriangleMesh&&) already forwards the mesh as an rvalue to ModelVolume. Before this change, ModelVolume did not have a matching TriangleMesh&& constructor, so the call selected the const TriangleMesh& constructor. That constructor creates a new TriangleMesh(mesh), copying the vertex and index buffers even though the rvalue no longer needs to retain its original storage.

The added constructor keeps the existing std::shared_ptr storage model, but moves the mesh into that shared object. This preserves immutable mesh storage, leaves the lvalue overload and other ownership behavior unchanged, and keeps the existing convex-hull initialization.

The regression test checks that the vertex and index buffer addresses are transferred rather than copied, and that the convex hull is still initialized. On high-poly models, this path can involve millions of vertices and indices, so the goal is to reduce loading memory traffic and peak memory without changing geometry or slicing behavior.

I chose this implementation to keep the change local to ModelVolume and avoid changing the existing ModelObject::add_volume API. If you prefer this to be tracked under a performance issue, or would like a dedicated benchmark added to the PR, I can add that and adjust the patch accordingly.

[Haidiye00]

approve