fix: use lmdb_get to access already loaded data without re-reading

[Robert-Forrest]

This small PR is a fix for the following kind of error:

File "~/mattergen/.venv/bin/mattergen-evaluate", line 10, in <module>
    sys.exit(_main())
File "~/mattergen/mattergen/scripts/evaluate.py", line 60, in _main
    fire.Fire(main)
File "~/mattergen/.venv/lib/python3.10/site-packages/fire/core.py", line 135, in Fire
    component_trace = _Fire(component, args, parsed_flag_args, context, name)
File "~/mattergen/.venv/lib/python3.10/site-packages/fire/core.py", line 468, in _Fire
    component, remaining_args = _CallAndUpdateTrace(
File "~/mattergen/.venv/lib/python3.10/site-packages/fire/core.py", line 684, in _CallAndUpdateTrace
    component = fn(*varargs, **kwargs)
File "~/mattergen/mattergen/scripts/evaluate.py", line 43, in main
    reference = LMDBGZSerializer().deserialize(reference_dataset_path)
File "~/mattergen/mattergen/evaluation/reference/reference_dataset_serializer.py", line 107, in deserialize
    impl=LMDBBackedReferenceDatasetImpl(lmdb_path, cleanup_dir=True),
File "~/mattergen/mattergen/evaluation/reference/reference_dataset_serializer.py", line 139, in __init__
    self._build_num_entries_by_chemsys_reduced_formulas(lmdb_path)
File "~/mattergen/mattergen/evaluation/reference/reference_dataset_serializer.py", line 150, in _build_num_entries_by_chemsys_reduced_formulas
    chemical_systems = lmdb_read_metadata(lmdb_path, "chemical_systems")
File "~/mattergen/mattergen/evaluation/utils/lmdb_utils.py", line 41, in lmdb_read_metadata
    with lmdb_open(db_path, readonly=True) as db:
File "~/mattergen/mattergen/evaluation/utils/lmdb_utils.py", line 21, in lmdb_open
    return lmdb.open(
lmdb.Error: The environment '****' is already open in this process.

which started occurring following the release of this change to LMDB: jnwatson/py-lmdb@2b26c9f, making opening the same environment more than once an error.

An alternative fix to this issue would be to pin the lmdb version used by mattergen to be <2.0.0, but then we remain vulnerable to the kinds of issues the change to lmdb aims to avoid (e.g. potential segfaults).

As far as I understand, Mattergen evaluation may be blocked in many environments currently, due to installation potentially picking up the latest lmdb version.

[Robert-Forrest]

@microsoft-github-policy-service agree [company="{your company}"]

@microsoft-github-policy-service agree company="Materials Nexus"