Remove synced-flush

[dnhatn]

To simplify rolling and full cluster restarts, we want to remove the need for a synced flush and allow quicker recoveries even if indexing is ongoing during the rolling restart. By moving everything to peer recovery retention leases, we can get rid of synced flush in 8.0, as a regular flush then has the same effect as a synced flush.

• Replace a synced-flush with a regular flush on shard inactivity (need to backport to 7.x) (Flush instead of synced-flush inactive shards #51365)
• Deprecate synced-flush in 7.x (Deprecate synced flush #50835)
• Adjust full cluster restart guideline in 8.0 (done in Goodbye and thank you synced flush! #50882)
• Remove synced-flush in 8.0 (Goodbye and thank you synced flush! #50882)
• Remove sync_id logic in RecoverySourceHandler (must keep it in 8.0)
• Remove syncedFlush logic in InternalEngine (Remove sync flush logic in Engine #51450)
  - [ ] Flush does not need to hold Engine#readLock?
• Stronger tests for full cluster restart with synced flush without peer recovery retention leases (Add test verify replica allocator with sync_id #51512)

[elasticmachine]

Pinging @elastic/es-distributed (:Distributed/Recovery)