Things to do to make the job cache more useful

• Datasets imported from a library are not recognized as being the same
• The input and output datasets need to be owned by the user -- accessible should be enough (though that might be a concern with our accessible by default policy)
• #6781 having the HID as part of the name really requires one to have the exact same history structure
• A tool property that says "ignore the element_identifier / name when caching" would be helpful
• A tool property to indicate the tool outputs are not reproducible (e.g. tools that produce random results on purpose)
• An additional setting that would allow the user to specify that an exact metadata match isn't necessary
• Enable cache override via UI element in the tool form.
• A preview
• Make datasets read-only in object store https://github.com/galaxyproject/galaxy/issues/8263
• If we have a hash for dataset contents we can "resume" cached execution when intermediate files get deleted or when the dataset metadata for an input file didn't match but the metadata mismatch didn't result in a different output
• Add something to the UI to indicate that a job was recovered from cached data
• https://github.com/galaxyproject/galaxy/issues/20050

No files, tests, or entry points are named. Start by reviewing the unchecked items and linked issues #6781, #8263, and #20050, then narrow the work to one agreed improvement with explicit completion criteria.