Content hashing design

[SeanTAllen]

Content hashing design decisions:

Hash the logical content, not the archive bytes. The hash is computed over the extracted files on disk, not the archive encoding. It survives format changes, re-encoding, or alternate transports.

Full Merkle tree with per-file proofs. SHA-256 throughout. Leaf nodes: SHA-256(relative_path_bytes || 0x00 || file_content_bytes) — binding the path into the leaf prevents swapped-content collisions. Interior nodes: SHA-256(left || right). Odd nodes promote without pairing; no duplication (duplicating the last node is a known weakness that enables second-preimage attacks).

Canonical ordering: leaf entries sorted by relative path as raw bytes, lexicographic. No locale sensitivity, no case folding.

Root hash is what goes in the config file and the ledger. Per-file membership is provable with the file's leaf hash plus sibling hashes to the root.

[SeanTAllen]

Implemented in #6000.