fix: remove dead restart_failed_sync from custody backfill

[ANtutov]

Custody backfill no longer uses the restart_failed_sync flag to gate restarts and never reads it, unlike the regular backfill logic. This change removes the unused field and its assignments from CustodyBackFillSync, drops the call from SyncManager, and leaves fully_synced_peer_joined as an explicit no-op for API symmetry

[cla-assistant[bot]]

All committers have signed the CLA.

[cla-assistant[bot]]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[eserilev]

we probably want to keep restart_failed_sync and actually use it correctly, instead of deleting it

[ANtutov]

Also squashed tp reduce noise in commits

[eserilev]

please read my comment above. We probably want to keep restard_failed_sync and fix its usage instead of deleting it

[ANtutov]

please read my comment above. We probably want to keep restard_failed_sync and fix its usage instead of deleting it

Done. I kept restart_failed_sync and wired it up like in BackFillSync, added a failed_sync flag so custody backfill only restarts after a failure once a new fully-synced peer joins.