Volume clipping tool with 3D rendering

[pskeshu]

Motivation

A virtual microtome — show the VLM arbitrary cross-sections through the embryo volume. This reveals internal 3D structure (body segment topology, coiling patterns) that projections collapse.

Design

This is more than numpy array slicing. The tool should produce rendered views similar to the viz server's 3D viewer (ray marching, alpha compositing, transfer functions), then clip to show cross-sections. This gives depth-aware rendered views, not raw optical sections.

The clipping tool could share rendering infrastructure with the frontend's existing 3D projection viewer (WebGL ray marching through a Data3DTexture).

Server-side rendering approach:

• Render the volume from an arbitrary viewpoint using alpha compositing
• Apply a clipping plane to expose internal structure
• Return the rendered image as base64 for the VLM

Relationship to z-slice analysis (#4)

Complementary, not redundant:

• Z-slices (Z-slice subagent for fold-stage discrimination #4) show raw optical sections — what the microscope actually captured
• Rendered cross-sections show depth-aware 3D structure with visual cues (shading, compositing) that help interpret spatial relationships

Integration

Could be a perception tool available to agentic experiments (experiments/tools/). The VLM requests a specific clipping plane and viewing angle, receives a rendered image.