#3238 datastore setup filesystem back to the default .swamp copies 0 files from the current datastore and exits 0, leaving the repo's data unreachable

What happens

After moving a repo's datastore to an external filesystem path, running swamp datastore setup filesystem to move it back to the repo's default .swamp directory switches .swamp.yaml to the new path but copies nothing from the current datastore. It reports success with filesCopied: 0, directoriesMigrated: [] and errors: [], and exits 0.

Afterwards swamp reads the nearly empty local .swamp, so all model data appears to be gone (swamp data query 'model("<name>")' returns 0 results) while the data still sits at the old external path. Nothing is deleted, but nothing warns that the data was left behind either.

Steps to reproduce

Swamp 20261009.215038.0, macOS. The external path here was an SMB mount, but any directory outside the repo should do.

# repo with some model data in the default .swamp, no datastore block in .swamp.yaml
swamp datastore setup filesystem --path /Volumes/NAS/test --json
# -> filesCopied: 131, directoriesMigrated: [definitions-evaluated, workflows-evaluated, data, outputs, workflow-runs, audit]
#    data is moved: local .swamp drops from 17M to 1.1M

swamp datastore setup filesystem --path "$PWD/.swamp" --json
# -> {"type":"filesystem","path":".../.swamp","filesCopied":0,"filesPulled":0,"bytesCopied":0,
#     "directoriesMigrated":[],"errors":[],"warnings":[]}   exit 0

swamp data query 'model("<name>")' --json   # -> 0 results
du -sh .swamp /Volumes/NAS/test             # -> 1.1M local, 17M still on the external path

.swamp.yaml now has datastore: { type: filesystem, path: $HOME/.../.swamp }, so swamp no longer looks at the external path.

Expected

Moving the datastore from any current location to any new location, including back to the default .swamp, should copy the current datastore's contents to the destination, the same as the first move did. Failing that, setup should refuse or warn that the existing data at the current path will not be migrated, rather than reporting success.

Workaround

Copy the data back by hand (rsync -a /Volumes/NAS/test/ .swamp/), restore .swamp.yaml to its original contents (no datastore block), then run swamp doctor datastores --repair -y. That step was needed because the catalog copied from the external datastore was missing rows for data written while the datastore was external (12 records on disk, 10 in the catalog).

Possibly related

#3162 (shipped): setup with --path .swamp in a repo that was never moved. If that fix skips migration whenever the destination resolves to the default .swamp, rather than comparing against the current datastore path, it would explain this behaviour. Not verified.