refactor(hot): delaminate sharding from HistoryRead/HistoryWrite

[prestwich]

Closes ENG-2287. Subsumes #66 (size-based shard splitter) — the splitter is included here as a backend-private detail of the MDBX impl.

Summary

Reshapes HistoryRead / HistoryWrite in signet-hot so that sharding is a backend-internal concern. The trait surface now exposes only logical history operations; ShardedKey<U256> no longer appears in any public method signature or return type.

Trait layout:

• crate::db::history::HistoryRead — blanket-impled on HotKvRead, default-impl-only. Backends cannot override (orphan rules + blanket impl structurally enforce this).
• crate::db::history::HistoryWrite — required per-backend, four methods: append_account_history, append_storage_history, truncate_account_history_above, truncate_storage_history_above. Plus default impls for bulk operations (update_history_indices, append_blocks, unwind_above, load_genesis, etc.).

Backend impls:

• MemKv writes a single dup entry per (addr) at u64::MAX (or per (addr, slot) at ShardedKey(slot, u64::MAX)). No splitter.
• signet-hot-mdbx uses the new signet_storage_types::merge_and_split with MAX_SHARD_BYTES = 1500 (constant private to the crate). Load-merge-rewrite for appends; backwards-walk delete-and-re-key-tail for truncates. The u64::MAX-tail subkey convention is preserved structurally.

Splitter (signet_storage_types::merge_and_split):

• Pure function: takes ownership of (existing, additions, max_bytes), returns (first, second) with second = Some(tail) iff a split occurred.
• Zero allocations on the no-split fast path beyond roaring container growth; exactly one IntegerList allocation on overflow.
• Worst-case roaring sizes pinned with property tests in signet-storage-types (dense pack of 650 contiguous values, sparse 100 distinct 16-bit containers, and round-trip through merge_and_split at the 1500 B budget).

Deleted:

• LegacyHistoryRead, LegacyUnsafeHistoryWrite, LegacyConsistentHistoryWrite.
• append_to_sharded_history free function in crates/hot/src/db/inconsistent.rs.
• ShardedKey::SHARD_COUNT constant (count-based splitting is obsolete).
• crates/hot/src/db/consistent.rs (contents moved as default impls onto HistoryWrite).

Design + plan docs: docs/superpowers/specs/2026-05-22-history-trait-delamination-design.md and docs/superpowers/plans/2026-05-22-history-trait-delamination.md (both gitignored; available locally on the branch).

Why now

PR #66 surfaced two layering concerns that this refactor resolves:

1. The size-based splitter in append_to_sharded_history referenced MAX_SHARD_BYTES (an MDBX-specific cap) from the abstract crate. The constant now lives in signet-hot-mdbx and the splitter is a pure helper in signet-storage-types.
2. ShardedKey<U256> appeared in public return types (last_storage_history), forcing every backend — even ones with no DUPSORT value cap — to mimic MDBX's sharded layout. The new logical interface lets each backend choose its strategy.

Test plan

• cargo +nightly fmt -- --check clean
• cargo clippy --workspace --all-targets --all-features -- -D warnings clean
• cargo clippy --workspace --all-targets --no-default-features -- -D warnings clean
• RUSTDOCFLAGS=\"-D warnings\" cargo doc --workspace --no-deps --all-features clean
• cargo t --workspace — 196 tests pass across all crates
• signet-hot conformance suite runs against both MemKv and MDBX backends (6 history conformance entrypoints now wired into conformance())
• signet-hot-mdbx structural test (history_sharding.rs) confirms MDBX splits oversized input into ≥2 dups and preserves the u64::MAX-tail subkey
• signet-hot integration tests (load_genesis for local, mainnet, parmigiana, pecorino genesis artifacts) pass
• signet-storage::unified end-to-end tests (append + drain_above + unwind) pass
• signet-hot-mdbx unwind conformance: byte-level comparison after unwind against a fresh-build replica
• Worst-case roaring size tests in signet-storage-types pin encoded bytes vs the 1500 B budget

Commit list (21 commits)

5b24c64 chore(hot): final cleanup from delamination review
2599ed9 test(hot-mdbx): structural assertion that MDBX splits oversized history
14d67de chore(storage-types): drop obsolete ShardedKey::SHARD_COUNT
28aedf8 refactor(hot): delete legacy history surface
02e96a0 test(hot): switch load_genesis integration test to logical history reads
5ba3949 test(hot): rewrite range conformance tests in logical terms
04739c7 test(hot): rewrite history conformance tests in logical terms
c6af236 refactor(hot): move non-shard-aware history helpers onto new HistoryWrite
d14f793 fix(storage): un-gate read-only UnifiedStorage methods from HistoryWrite
c186c8f refactor(hot): move consistent history ops onto new HistoryWrite
6558eec refactor(hot): flip revm test setup to logical HistoryWrite
bac8871 feat(hot-mdbx): implement HistoryWrite using merge_and_split
900a567 feat(hot): implement HistoryWrite for MemKv
aaa27f8 fixup(hot): panic on impossible-state in blocks_changed_* default impls
acc37bd feat(hot): introduce logical HistoryRead / HistoryWrite trait module
0a0bd2d refactor(hot): rename legacy history traits in preparation for delamination
0b54bf4 test(storage-types): pin BlockNumberList worst-case sizes against MDBX budget
e77c2e6 fixup(storage-types): restore debug_assert on merge_and_split precondition
a27e25a feat(storage-types): add merge_and_split for size-bounded history shards
f40f7a4 feat(storage-types): add BlockNumberList::pop_max

Note: commits are unsigned in this draft; they will be re-signed via git rebase --exec 'git commit --amend --no-edit -S' before merge.

🤖 Generated with Claude Code

[prestwich]

rustdoc should reference present behavior only, not past behavior

[prestwich]

this should be a function on IntegerList.
and the additions should not be materialized

[prestwich]

why was this file changed?

[prestwich]

Remove this section of rustdoc. item-level rustdoc describes ONLY the behavior and how the downstream user interacts with it. archietecture goes in module or lib level