sanitizeOutput still hashes the CWD for empty strings in 1.1.1

Summary

sanitizeOutput() / OutputSanitizer.recursiveParse treats every String as a filesystem path. In nft-utils 1.1.1 (latest release) an empty string becomes the current working directory:

java.nio.file.Path path = Paths.get(strValue);
if (Files.isDirectory(path)) {
    // recursively list + hash children
}

Paths.get("") is the CWD, Files.isDirectory is true, and the sanitizer walks the whole tree (.git, plugin caches, conda/pixi envs, …). Nested copies of that listing then explode workflow snapshots.

Reproducer

Any sanitizeOutput() of a map/channel that contains "", for example an unset BAM @RG CN stored as an empty string:

assert snapshot(sanitizeOutput([meta: [CN: '']])).match()

In nf-cmgg/preprocessing this turned tests/workflows/preprocessing.nf.test.snap into a ~27 MB file. readgroup.CN was a recursive MD5 listing of the repo, repeated on every FASTP emit that still carried sample meta.

Expected

Leave empty / non-path strings alone. Only walk a value when it is actually an output directory.

Related

This is the same root cause as #84. The empty-string guard landed on main in #85 (if (strValue.isEmpty()) return IGNORE) after 1.1.1 and has not been released.

Please cut a 1.1.2 that includes #85.

Remaining gap on main

#85 only skips empty strings. Any other relative directory name still matches Files.isDirectory and is walked, including ".", "..", and names that happen to exist in the test CWD ("docs", "conf", …). A safer check would require an absolute path (or otherwise look like a pipeline output path) before recursing.

Start at OutputSanitizer.recursiveParse and reproduce the empty-string case with the sanitizeOutput assertion shown in the issue. Check the existing empty-string handling from #85 and define the remaining path cases to cover; done means empty and non-path strings are left alone and output directories are sanitized without recursively walking unrelated directories.