feat(cold): add stream_logs for incremental log streaming

[prestwich]

Summary

• Add stream_logs method to the ColdStorage trait for incremental log streaming via bounded mpsc channels, avoiding full result materialization into Vec<T>
• Implement for all three backends: in-memory (MemColdBackend), MDBX (MdbxColdBackend), and SQL (SqlColdBackend — SQLite + PostgreSQL)
• Add StreamLogs request variant, handle method, and runner dispatch for the task-based architecture
• Add streaming conformance tests and two new error variants (StreamDeadlineExceeded, ReorgDetected)
• Bump workspace version to 0.6.0 (new required trait method)
• Make DatabaseEnv cheaply cloneable (derive Clone)
• Eliminate per-row Vec<u8> heap allocations in SQL row extraction (borrow &[u8] from rows)
• Borrow filter params as &[u8] from the Filter struct instead of cloning into Vec<Vec<u8>>
• Remove get_block_hash and get_logs_block from the ColdStorage trait (replaced by existing get_header/get_logs)
• Fix TooManyLogs bug: MDBX and default streaming reported remaining instead of max_logs in error, causing wrong limit values when logs span multiple blocks
• Introduce StreamParams struct: bundles from, to, max_logs, sender, deadline to reduce produce_log_stream from 7 parameters to 2
• Add try_stream! macros: reduce repeated match/send-error/return boilerplate in MDBX and SQL streaming implementations
• Borrow Filter in get_logs: change from filter: Filter to filter: &Filter, eliminating per-block clone in the default streaming helper (N clones → 1)

Design

Each stream_logs call spawns a producer task that sends RpcLog items through a bounded mpsc channel. The consumer receives a ReceiverStream<ColdResult<RpcLog>> (concrete type, no boxing).

Key properties:

• Active deadline enforcement — producer checks deadline at block boundaries and uses timeout_at on channel sends, guaranteeing resource release within stream_deadline regardless of consumer behavior
• MVCC snapshot consistency — MDBX uses a single read transaction via spawn_blocking; PostgreSQL uses REPEATABLE READ isolation
• Reorg detection — default implementation anchors to the to block hash at stream start, re-checked before every block; snapshot backends skip this (unnecessary under MVCC)
• Concurrent stream limiting — backend-owned semaphore (8 permits), held by the producer task and released on exit
• Clean cancellation — dropping the stream causes send to fail, producer exits immediately
• Minimal allocations — SQL row extraction borrows &[u8] from rows; filter params borrow from the Filter struct; DatabaseEnv clones cheaply without Arc wrapper; get_logs borrows &Filter avoiding ownership transfer

Test plan

• cargo clippy -p signet-cold --all-features --all-targets
• cargo clippy -p signet-cold --no-default-features --all-targets
• cargo clippy -p signet-cold-mdbx --all-features --all-targets
• cargo clippy -p signet-cold-mdbx --no-default-features --all-targets
• cargo clippy -p signet-cold-sql --all-features --all-targets
• cargo clippy -p signet-cold-sql --no-default-features --all-targets
• cargo clippy -p signet-hot-mdbx --all-features --all-targets
• cargo clippy -p signet-hot-mdbx --no-default-features --all-targets
• cargo +nightly fmt
• cargo t -p signet-hot-mdbx (44 passed)
• cargo t -p signet-cold (2 passed)
• cargo t -p signet-cold-mdbx (4 passed)
• cargo t -p signet-cold-sql --all-features (2 passed — SQLite + PG conformance)
• ./scripts/test-postgres.sh (PostgreSQL conformance via Docker)

🤖 Generated with Claude Code

[prestwich]

generally review for unnecessary allocations

[prestwich]

[Claude Code]

Addressed all review comments in 2e3b4f1:

1. Functional rewrite of collect_logs_block (comment 1) — Collect the fallible MDBX iterator first, then flat_map → filter_map → collect for the inner log iteration.

2. Extract check_block_hash helper (comment 2) — Flattens 3 levels of match nesting into a ? chain. Also serves as the MDBX get_block_hash trait impl.

3. DRY streaming into shared task (comments 3 + 6) — Removed stream_logs from the ColdStorage trait entirely. Replaced with two simpler primitives (get_block_hash, get_logs_block). The streaming loop (semaphore, deadline, reorg detection, channel sends) now lives solely in ColdStorageTaskInner::handle_stream_logs. Eliminates ~200 lines of duplicated code across 3 backends.

4. Extract append_filter_clause utility (comments 4 + 5) — Deduplicated the single-vs-multi-value SQL clause pattern into a shared function. build_log_filter_clause now calls it for both address and topic filters.

5. Remove unnecessary collect (comment 7) — SQL get_logs_block checks rows.len() > remaining before mapping, avoiding an intermediate allocation.

6. Allocation review (general comment) — Removed unused tokio and tokio-stream dependencies from both backend crates. Conformance tests now run through ColdStorageHandle (spawning a ColdStorageTask internally).

All clippy (both --all-features and --no-default-features), tests, docs, and test-postgres.sh pass.