The transporter should properly handle symlinks it encounters in the task's work directory when uploading outputs.
If the symlink is to an input, skip uploading the output and instead use the original input's URL.
If the symlink is to another file within the work directory, skip uploading the file and use the URL of the previously uploaded file (symlinks should be handled after files).
If the symlink is to a file that is not within a work directory, return an error.
See also stjude-rust-labs/sprocket#580.