feat(specs,test-types): add binary tree state provider behind a --state-trie flag

[spencer-tb]

Description

Upstreams the commitment-neutral infrastructure of projects/binary-trie — and deliberately nothing the EIP process governs.

The pain of long-lived project branches is rebasing shared framework and tooling churn and running a separate release pipeline — not carrying the spec content, which is small and stable. This PR moves exactly that shared tier, and keeps the line the EIP process draws:

• Zero changes under src/ethereum/forks/. No fork copy, no pseudo-fork, no EIP-8297 semantics in any fork. The spec's fork lineage (and its rendered docs) remains exactly the set of scheduled forks; a binary_tree fork materializes if and when the EIP process schedules 8297, so repo structure keeps tracking governance rather than front-running it.
• The tree is shared infrastructure beside the MPT — partitioned_binary_tree.py ↔ merkle_patricia_trie.py, state_pbt.py ↔ state_mpt.py — selected per session by a testing flag. A tooling property, not a protocol claim: every fork defaults to MPT.
• Release integration is a bonus tarball beside existing release assets that can never block a release.

What stays on projects/binary-trie because the EIP process owns it: the fork copy and its parity test, the tests/binary_tree/ EIP-8297 fixture suite, test_spec_constants.py, and the cross-client conformance-vectors tooling.

Contents by commit

• feat(specs,test-types): ethereum/partitioned_binary_tree.py (structure + embedding, one module) and ethereum/state_pbt.py (PreState provider mirroring state_mpt.py); error contracts documented on the PreState protocol (get_code raises UnknownCodeHashError on malformed pre-state — state_mpt updated to match; compute_state_root raises InvalidBlock for state the commitment cannot encode, which the t8n reports as a rejected block with the pre-state root); StateCommitment.PBT + Alloc routing; blake3 dep; tests/binary_trie/ unit suites (fill-ignored) incl. a differential-vs-MPT suite and a test driving Amsterdam's unmodified execute_block against a binary-tree state (call sites type-ignored until the pre_state annotation is loosened; projects/zkevm carries that loosening). 100% line and branch coverage of both new modules.
• feat(test-forks) (carried from feat(test-forks): add --state-trie state-commitment override flag marioevz/execution-specs#9): --state-trie {mpt,pbt} session override of Fork.state_commitment(); pytest_unconfigure restores the default so in-process pytester sessions cannot leak it; fork-agnostic tests for defaults, override incl. transition forks, option wiring, and genesis/pre-alloc-group behavior under the override.
• feat(ci): binary-trie job — unit suites with coverage plus a full-tree PBT-mode fill (just fill-state-trie-pbt); the fill lane doubles as the enumeration of EIP-8297-incompatible tests (balances ≥ 2^128) until those are curated.
• feat(ci): release variants — tests@ and <feat>-devnet@ releases attach <tarball>_binary.tar.gz, filled Amsterdam-only in PBT mode on its own runner through the standard build/combine pipeline (variants: in feature.yaml); a failed variant fill never blocks the release. A follow-up PR adds a zkevm variant on the same mechanism.

Verified: default fills byte-identical to base (modulo the _info.url commit pin); fill --fork Amsterdam --state-trie pbt produces binary state roots end to end; just static clean.

Related Issues or PRs

N/A.

Checklist

• Ran fast static checks to avoid CI fails, see Code Standards & Verifying Changes: just static
• PR title has the form <type>(<area>): <title>, where <type> and <area> come from an appropriate C-<type>, respectively A-<area>, label. The title should match the target squash commit message.

Cute Animal Picture