[MeABzZz]

Optional: decision for reviewer. Draft. Companion to BTCDecoded/blvm-consensus#24 (assume-valid with Core semantics). This PR doesn't compile against published blvm-consensus until #24 is released, because it sets the new assume_valid_ancestor field. Locally it builds and passes through the sibling [patch.crates-io] paths.

What it does

BlockStore::assume_valid_ancestor_lookup() returns the hook consensus needs: the assume-valid block's ancestor at height h on the stored header tree (Core: assumevalid->GetAncestor(h)).

• The ancestry is built once per process: walk prev_block_hash back from the assume-valid header, which gives one 32-byte hash per height. That's about 30 MB on mainnet. It's cached in a static, keyed by (hash, height).
• It doesn't trust the height index. That index is first-seen per height, so a competing header chain seen first would otherwise make its blocks look like ancestors and skip scripts.
• Until the assume-valid header is stored, the lookup returns None (nothing skips) and retries cheaply: one header read, no allocation.

Wired next to difficulty_ancestor on these paths:

• parallel IBD (parallel_ibd/mod.rs)
• the sequential connect in block_processor.rs
• reorg_executor.rs
• pruning.rs

The blvm-protocol validate_and_connect_block_with_difficulty paths (block_processor's first pass, RPC) build their own context and still verify every script. That's fail-closed, and those paths already verify fully today in block_processor.

Test: tests/assume_valid_ancestor_lookup.rs

Store a competing header chain for heights 0..=3 first, so it occupies the height index, then the real 10-header chain with assume-valid at height 6. Checks:

• before the real chain arrives: None
• after: heights 0..=6 return the real chain's hashes, even where the height index still holds the competing hash
• height 7: None

Also green: fork_reorg_regtest, fork_reorg_events, blockchain_rpc_tests, cargo test --lib parallel_ibd (727 passed), and cargo clippy --lib with no new warnings.

Not done: an IBD-throughput measurement with the hook on mainnet headers. The one-time walk is about 938k header reads at the first block below the assume-valid height.

Merge order

1. Optional: assume-valid with Core semantics (skip only ancestors, never reject) blvm-consensus#24, then a consensus release.
2. This PR, with node's blvm-consensus pin bumped to that release.

Relative to the 7 criticals PRs: works with or without #31 (Core v31.1 hash and height per network). With #31, the default hash is anchored, so this lookup finds it once headers sync. No file overlap with #31, #32 or #38. #38 also touches reorg_executor.rs, but in a different hunk.