v0.13.1

Features

• Sprite rendering: place camera-facing textured billboards in 3D space via SpriteItem. Per-sprite controls include position, color, size, rotation, and UV rect. Size can be screen-space (pixels, constant on screen regardless of distance) or world-space (expands along the camera right/up vectors so sprites scale with distance). Set depth_write to false for additive particle effects that composite correctly against opaque geometry, or true for opaque markers. Use uv_rects to sample subregions of an atlas texture and cycle frames each tick for sprite-sheet animation. Particle simulation stays in host code; push positions and per-particle colors/sizes each frame.
• VolumeSurfaceSliceItem: samples a 3D volume on an arbitrary surface mesh and colors each fragment by the volume scalar at that world position. Unlike ImageSliceItem, the slice surface is not restricted to axis-aligned quads -- any mesh produced by upload_mesh_data works: flat planes, disks, saddle surfaces, paraboloids, or imported geometry. Fragments whose world position falls outside the volume bounding box are discarded automatically. Push a VolumeSurfaceSliceItem into SceneFrame::volume_surface_slices each frame, referencing the mesh by MeshId and the volume by VolumeId.
• prepare_callback / paint_callback: unified entry points for the eframe CallbackTrait model. Call prepare_callback from CallbackTrait::prepare and paint_callback from CallbackTrait::paint instead of managing separate prepare, prepare_ldr_dyn_res, and prepare_hdr_callback calls manually. The methods dispatch internally based on whether post_process.enabled is set, so OIT, EDL, and tone-mapping are active in eframe apps without extra code in the callback.
• TransparentVolumeMeshItem gains threshold_min and threshold_max fields. Tetrahedra whose scalar value falls outside the range are discarded by the shader without re-uploading geometry. Set both fields to the same raw scalar units used at upload time. Defaults to no clipping (all tets rendered).
• ViewportGpuResources::replace_sparse_volume_grid_data: replaces a previously uploaded sparse voxel grid in place without allocating a new mesh slot. Use this for per-interaction updates such as voxel painting to avoid leaking GPU memory. Mirrors the existing replace_clipped_volume_mesh_data pattern.

Fixes

• Uploading a large transparent volume mesh could crash with an out-of-memory error. The decomposed tet data is now split into device-limit-bounded chunks on upload, so meshes of any size load correctly. No API changes.
• Transparent volume meshes and Surface LIC were silently invisible in eframe applications. Both features require the HDR pipeline, which was never activated via the prepare + paint callback path.