What you want is only possible with filesystem-level backup. Index entries ultimately contain references to tuple IDs (TID) of table rows, which are essentially physical pointers to records within database files. Every row update will move it to a different location with a different TID, making the old TID useless unless updated concurrently.
A table restored from a dump will have (possibly all) row TIDs different from the original; if you somehow managed to duplicate the original index and use it to access this new table, results would be undefined.
Only a filesystem-level backup of the index files together with the corresponding table file(s) can guarantee that TIDs remain unchanged.