safekeeper: Fix local_start_lsn on sk4

We're seeing errors on safekeeper-4 in us-east-2, and they are localized only on this safekeeper.

2024-01-23T15:12:42.309924Z ERROR {cid=12138 ttid=XXX/YYY}:WAL sender: terminated: Other(Failed to open WAL segment download stream for remote path RemotePath("XXX/YYY/000000010000000000000003")
Caused by:
    No file found for the remote object id given
Stack backtrace:

safekeeper-4 is the only safekeeper that has timelines with local_start_lsn != timeline_start_lsn. This is quite possibly causing the issues here, as this error can happen if client (pageserver) requests WAL from the same segment as where local_start_lsn is located, but before local_start_lsn itself. The logic in safekeepers prevents reading uninitialized WAL and safekeeper is trying to read WAL from remote storage, but it can be unavailable if this segments wasn't uploaded yet.

The plan is:

implement https://github.com/neondatabase/neon/issues/6397
collect a list of timelines with local_start_lsn != timeline_start_lsn
verify that hash of segments are the same as on other safekeepers
use #6397 to update local_start_lsn

Related slack threads:

neondatabase / neon

safekeeper: Fix local_start_lsn on sk4 #6449