feat(shader-driver): migrate FingerprintColumns.cycle to Vsa16kF32 carrier (PR B)

[AdaWorldAPI]

Summary

• PR B from the SoAReview expansion list §6, item feat: bump arrow 57, datafusion 51, lance 2 #1 — migrates FingerprintColumns.cycle from Box<[u64]> (Binary16K, 256 × u64 per row) to Box<[f32]> (Vsa16kF32 carrier, 16,384 × f32 per row = 64 KB).
• Builds on PR feat(contract): Vsa16kF32 switchboard carrier + 16K algebra #253 which shipped the CrystalFingerprint::Vsa16kF32 variant + 6 algebra primitives (vsa16k_zero, binary16k_to_vsa16k_bipolar, vsa16k_to_binary16k_threshold, vsa16k_bind, vsa16k_bundle, vsa16k_cosine).
• Upstream API unchanged — write_cycle_fingerprint() still takes &[u64; 256]; the bipolar projection happens internally via binary16k_to_vsa16k_bipolar. Downstream readers get &[f32] for direct VSA algebra.

What changed

File	Change
crates/cognitive-shader-driver/src/bindspace.rs	cycle: Box<[u64]> → Box<[f32]>; new FLOATS_PER_VSA = 16_384 const; set_cycle(&[f32]), set_cycle_from_bits(&[u64; 256]), cycle_row() -> &[f32]; byte_footprint() now 71,774 for 1 row
crates/cognitive-shader-driver/tests/end_to_end.rs	Test adapted — verifies bipolar round-trip via vsa16k_to_binary16k_threshold
.claude/board/LATEST_STATE.md	Cycle substrate migration noted

Why

Expansion-list item #1: the Vsa16kF32 carrier type exists (PR #253), but the BindSpace column it was designed for still used Binary16K. This PR closes that gap — FingerprintColumns.cycle now carries the 64 KB f32 carrier, enabling future operations (MarkovBundler, Trajectory, vsa_permute) to do VSA algebra directly on the cycle column instead of converting at every call site.

Test plan

• cargo test -p cognitive-shader-driver --lib — 40 pass, 0 fail
• cargo test -p lance-graph-contract --lib — no regressions
• byte_footprint() for 1 row = 71,774 bytes (3 × 256 × u64 + 16,384 × f32 + qualia + meta + edge + temporal + expert)
• Bipolar projection round-trip is lossless (strict ±1 → threshold inverse)

https://claude.ai/code/session_01SbYsmmbPf9YQuYbHZN52Zh

---

Generated by Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 72663523b3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid per-row heap allocation in cycle projection

set_cycle_from_bits currently calls binary16k_to_vsa16k_bipolar(bits), which allocates a new 16,384-f32 Box and then immediately copies it into the preallocated row buffer. Because write_cycle_fingerprint routes every persisted cycle through this path, this introduces allocator churn and an extra 64 KB copy per row in a hot path, which can materially degrade throughput/latency under sustained dispatch workloads.

Useful? React with 👍 / 👎.