Use oaklib shared cache instead of local .db files in data/raw/

[turbomam]

Problem

The transform step generates large semsql SQLite databases next to the downloaded .owl files in data/raw/:

• chebi.db (~3.4 GB)
• ncbitaxon.db (~11 GB)
• go.db (~385 MB)

This happens because constants.py defines sources as .owl paths and transform code calls get_adapter(f"sqlite:{CHEBI_SOURCE}"), which triggers oaklib to build the .db in-place via semsql.

These files duplicate what oaklib already caches at ~/.data/oaklib/ when using the sqlite:obo: scheme.

Proposed Fix

In transform scripts (e.g., rhea_mappings.py), change:

get_adapter(f"sqlite:{CHEBI_SOURCE}")  # builds data/raw/chebi.db from .owl

to:

get_adapter("sqlite:obo:chebi")  # uses ~/.data/oaklib/chebi.db (pre-built, cached)

This avoids ~15 GB of duplicated ontology databases per checkout and uses fresher pre-built databases from the S3 distribution.

Impact

• Saves ~15 GB disk per working copy
• Faster builds (skip semsql compilation of large ontologies)
• Shared cache benefits other oaklib-dependent projects on the same machine