If bottomless misses uploading a single frame, then restoration will fail

Lets say there is an network issue and bottomless failed to upload a single frame, then sent rest of the frames, then restoration can fail. Snapshot at shutdown fixes this issue, but if the DB did not have a clean shutdown then it won't work.

To reproduce this issue (you can use Nemesis):

1. Run sqld and insert a few rows in a db
2. Stop local minio server
3. Insert a few more rows
4. Resume minio and insert few more rows
5. Kill the sqld (i.e. unclean shutdown)
6. On restart, the restoration would fail

Possible solution:

We write all the frames to disk before uploading to S3. We could probably build a local queue and when an upload is successful, it should check if there are any previously unsent frames present on disk and upload them.

Start with the bottomless frame upload, local-disk queue, and restoration paths; the issue does not name specific files or tests. Reproduce the failure with the listed MinIO outage and unclean shutdown steps, then verify that a missed frame is uploaded later and restoration succeeds.