Resume cache and linkable-aware duplicate handling for upload_files

[Caushi]

• Add a resume_cache path to upload_files, keeping each file's tus URL keyed by MD5 so an interrupted transfer resumes instead of restarting from byte 0. None (the default) keeps no state and never resumes.
• Change on_duplicate="link" to link only to a duplicate the server reports as linkable; one whose resource is still pending or failed is uploaded instead, so a retry after a failed transfer no longer fails until B-Fabric's orphan cleanup runs (~2 days).
• Add linkable and existingResourceStatus to DuplicateResult. Breaking: on_duplicate="link" now requires a B-Fabric that reports linkable; a response omitting it is refused rather than guessed.
• Raise WorkunitCompletionError when every transfer succeeded but marking the workunit available failed, carrying the UploadSummary so a caller can retry only the status flip instead of creating a second workunit.