Preserve URI scheme when resolving Xenium bundle child paths

[rnaidu-seqera]

Problem

Running the pipeline on an object-storage work directory (S3 on AWS Batch) aborts during Xenium bundle validation, even though the UNTAR step succeeds:

❌ Xenium bundle does not exist: /<bucket>/work/7a/a1c7ff.../test_run
java.nio.file.NoSuchFileException: /<bucket>/work/.../test_run/gene_panel.json

Note the paths have no s3:// scheme — the bucket name became the first segment of a local path.

Root cause

Bundle child paths were built from Path.toString(). For remote filesystems, nextflow.cloud.aws.nio.S3Path.toString() returns /<bucket>/<key> and drops the s3:// scheme. The scheme-less string was then re-parsed with file(...), which resolved it on the local filesystem of the head node, where the files don't exist.

This is invisible on local / shared-filesystem executors (e.g. nf-core CI), because there toString() already yields a valid local path. It only manifests on object storage.

Affected sites in workflows/spatialaxe.nf:

• UNTAR output staging (-profile test)
• required/optional bundle-file checks (affects production runs too)
• transcripts.parquet, experiment.xenium, gene_panel.json, and the morphology-focus image resolution

Fix

Use scheme-preserving APIs instead of toString():

• Path.resolve(child) for the required/optional file existence checks (keeps the filesystem provider);
• bundle.toUriString() / file(bundle).toUriString() for the UNTAR staging value and the string-concatenation sites.

No behaviour change on local / shared-filesystem executors.

Validation

• Each rebuilt path resolves to the correct provider: S3Path for s3:// inputs (including the trailing-slash morphology_focus/ directory) and UnixPath for local inputs.
• nextflow lint passes with no new errors.
• A minimal reproducer confirms the failure on an S3 work dir before the change and success after.

PR checklist

• CHANGELOG.md is updated.
• Documentation update (n/a — no user-facing parameter/behaviour change).
• nf-test / CI green (please run in nf-core CI).

[heylf]

LGTM. Thank you so much @rnaidu-seqera for your fix and work ❤️