sanity_check_attributes bodies are mostly empty across visuals

[jeromeetienne]

Short tracking issue — most sanity_check_attributes / sanity_check_attributes_buffer static methods across the visuals/geometry/material/texture classes are either pass or limited to isinstance(x, Buffer) checks. As a result, malformed input (wrong dtype, wrong shape, mismatched counts) crashes deep in numpy at render time instead of failing clearly at construction or at the buffer boundary.

Audit (as of mesh_ng):

• gsp/visuals/mesh.py — sanity_check_attributes is pass; _buffer form only delegates.
• gsp/visuals/markers.py — both pass.
• gsp/visuals/paths.py — both pass.
• gsp/geometry/mesh_geometry.py — both pass.
• gsp/visuals/pixels.py, points.py, segments.py, texts.py — _buffer only has shallow isinstance(_, Buffer) checks; no shape/dtype/count validation.
• gsp/visuals/image.py, gsp/core/texture.py — partial type checks only.
• gsp/materials/mesh_basic_material.py — _buffer now enforces rgba8 on face/edge colors (Matplotlib mesh: 6 confirmed bugs + 2 to verify (vs. matplotlib_scenegraph) #20 Bug 5); the non-buffer form is still pass.

What good would look like (per class):

• Buffer BufferType matches the expected encoding (e.g. positions → vec3, indices → uint32, colors → rgba8).
• Buffer count consistency between related attributes (e.g. material color count ∈ {1, vertex_count, face_count}).
• Optionally O(N) range checks (e.g. max(indices) < vertex_count) at the buffer boundary.

Related: this is the larger Bug 6 from #20, broadened beyond Mesh.

This is just a placeholder to come back to — no fix scoped yet.