I can definitely see a use for checking the integrity of the decrypted/decompressed data. Of course, implementation bugs or hardware defects could wreck your data after that stage, but increasing coverage can still help.
Verifying data integrity isn't exactly the same as authenticating a message, so I think you'd probably want to use a simpler scheme for that. For example, a basic CRC would suffice for most use cases there.
Of course, if the compression/encryption method has some way of checking the integrity of the output (that's of equal strength to the signature) then then signing first would be completely redundant.
EDIT: so in many scenarios, signing first and last would have no advantage. For example if you get to decide what implementation will be used by the sender and the recipient (most package managers?)