v1.4.0 — embedded-Nano manifest backend + topological view ordering

Highlights

Embedded HeliosDB-Nano as an optional manifest backend

[options] manifest_backend = "nano" runs a2h's resumable-load ledger on an in-process HeliosDB-Nano (RocksDB) store instead of the stdlib sqlite3 default. SQLite stays the zero-friction default; nano is opt-in:

pip install 'any2heliosdb[nano-manifest]'   # adds heliosdb-nano-embedded >= 3.60.8

RW handles share one refcounted RocksDB writer (single-writer) behind a lock so the multi-threaded loader is safe; status/monitor open read-only and reopen per tick. Manifest SQL is portable across both backends. Validated with a real PG→PG migrate and a 1.2 M-row mid-load SIGKILL + a2h resume that recovered with zero duplicate rows.

Topological view ordering (export + migrate)

Views are emitted in dependency order, so a view selecting from another view is created after its referent (PostgreSQL and HeliosDB both require that). Self/recursive views aren't a self-dependency; cycles keep original order; independent views stay in source order.

Compatibility

Default behavior unchanged — SQLite manifest, no new required dependency. Full suite 256 passed / 10 skipped.

See CHANGELOG for details.