Enhancement: Report permanent/fatal backend errors earlier

When encountering errors in reading from or writing to storage backends,
restic retries the failing operation up to nine times (for a total of ten
attempts). It used to retry all backend operations, but now detects some
permanent error conditions so that it can report fatal errors earlier.

Permanent failures include local disks being full, SSH connections
dropping and permission errors.

https://github.com/restic/restic/issues/2453
https://github.com/restic/restic/pull/3170
https://github.com/restic/restic/issues/3180
https://github.com/restic/restic/pull/3181