Add readable multi-model filesystem vector views

[isomorphisms]

What this makes real

• Adds the requested views/retrieved-from-the-web/{text,images} and views/organizing-the-information/{categories,vector-spaces} layout.
• Implements an exact C99 vector backend whose authoritative vectors are fixed-width plain text: 15 ASCII bytes per signed scientific Float32 plus one delimiter byte. The byte offset is (row * dimensions + coordinate) * 16.
• Keeps the little-endian Float32 file only as a rebuildable speed cache. query-text works without it, compile-cache recreates it, column performs direct coordinate seeks, and check proves text/cache equality.
• Pins two deliberately different Hugging Face views by immutable commit, size, SHA-256, preprocessing, and output dimension:
  • minishlab/potion-base-2M ONNX: 64 coordinates, static lookup/mean/normalize.
  • mixedbread-ai/mxbai-embed-xsmall-v1 INT8 ONNX: 384 coordinates, transformer/mean/normalize.
• Adds verified fetch/materialization commands and copies each exact model manifest beside its vector-space collections.
• Runs both actual ONNX models through embed → filesystem build → cached query → text-only query in CI.

Verification

• Warning-clean C99 build and smoke suite.
• AddressSanitizer and UndefinedBehaviorSanitizer pass (leak detection disabled because the work sandbox runs under ptrace).
• Corrupt text, corrupt cache, missing cache, strict-manifest, empty-index, dimension, zero-norm, and score-overflow cases covered.
• 10,000 × 384 scale run produced exactly 61,440,000 readable bytes and 15,360,000 cache bytes; cached and text-only rankings were byte-identical.
• A distinct semantic query about closed nondegenerate two-forms retrieved the symplectic-geometry fixture first with both models.

Boundary

This completes the filesystem/index/model-provenance layer without a vector database. The ONNX adapter is deliberately under experiments/; moving selected inference into the phone-resident Android browsing loop remains a separate runtime step.

Stacked on #24.