Add LMDB storage backend, OCaml 5.5 Cloud Agent environment

[tiensonqin]

Summary

• Add datascript_lmdb: LMDB-backed implementation of the existing storage KV interface (address → Transit-encoded PSS node payload), mirroring the SQLite backend shape.
• Add persistent_storage_bench: runs the same snapshot + conn workloads against SQLite and LMDB side by side.
• Add test_lmdb_package: roundtrip + closed-session guard tests.
• Extend Cloud Agent / CI setup with liblmdb-dev and lmdb opam dependency.

Benchmark highlights (native OCaml 5.5)

Lower is better (ms). Same row counts for both backends; LMDB files are larger due to mmap pre-allocation.

Size 100

Operation	SQLite	LMDB
snapshot-build-and-store	3.85	1.90
snapshot-restore	0.06	0.02
snapshot-add-one-and-store	11.23	2.78
snapshot-update-and-store	7.14	2.35
conn-add-one	0.70	0.43
conn-update-one	0.36	0.26
file size after build	106 KB	197 KB

Size 1000

Operation	SQLite	LMDB
snapshot-build-and-store	30.83	31.92
snapshot-update-and-store	23.83	5.22
conn-update-one	0.95	1.11
file size after build	967 KB	1.8 MB

LMDB wins clearly on incremental snapshot store/update paths; bulk build is similar at 1000 blocks. SQLite produces smaller on-disk files.

Usage

opam exec -- dune exec bench/persistent_storage_bench.exe -- --size 100
opam exec -- dune exec bench/persistent_storage_bench.exe -- --sizes 100,1000,5000

let session = Datascript_lmdb.open_session path in
let storage = Datascript_lmdb.storage session in
(* same store / restore / create_conn API as SQLite *)

Cloud Agent environment

.cursor/environment.json pins OCaml 5.5.0 via Dockerfile + install script.