Research-grade verification foundation for v0.3

[JROChub]

Summary

• enforce prime-field construction and harden sparse decoders against allocation abuse
• add canonical cross-language conformance vectors, property tests, and exhaustive single-byte mutation rejection
• add reproducible soundness-budget and benchmark tools
• document the exact v1 security boundary, primary prior art, and falsifiable v2 research target

Key finding

The v1 sparse workflow is deterministic full replay of a public polynomial whose Boolean sum has a closed form. It is a strong reproducibility artifact, but not a succinct delegated-computation protocol. A conventional one-repetition million-round sum-check over the current field would have only about 9.97 bits under the classical n/|F| bound. The proposed v2 target uses a larger field/repetition budget and a reviewed commitment/opening layer.

Validation

• cargo fmt --check
• cargo clippy --all-targets --all-features -- -D warnings
• cargo test --all-targets --locked (42 + 6 tests)
• cargo test --all-targets --features net --locked (57 + 6 tests)
• python3 scripts/test_sparse_verifier.py
• cargo package --allow-dirty
• full published million-round Rust/Python artifact replay