Optimize globe satellite rendering with InstancedMesh matrix buffer updates

[seanwevans]

Motivation

• Reduce CPU/GPU overhead when rendering large numbers of satellites by batching transforms into a single instanced draw path.
• Support up to 100_000 objects so the globe can render high-density streams without per-instance draw calls.

Description

• Increase instance capacity by changing MAX_OBJECTS from 10_000 to 100_000 and add MATRIX_SIZE and SATELLITE_SCALE constants.
• Import Quaternion and Vector3 and add a matrixBufferRef that stores a direct reference to InstancedMesh.instanceMatrix.array for typed-array updates.
• Replace per-instance setMatrixAt calls with composing a transform (position, rotation, scale) and writing each 4x4 matrix directly into the shared Float32Array via matrix.toArray(matrixBuffer, i * MATRIX_SIZE).
• After batched writes, toggle instanced.instanceMatrix.needsUpdate = true, and clear the buffer reference during cleanup to avoid stale pointers.

Testing

• Ran the production build with npm --prefix web run build, which completed successfully (TypeScript compile + Vite build).

---

Codex Task