Join of a snapshot-less node to an original node replays the ledger

[cjen1-msft]

When a node tries to join, it sends a startup_seqno to the primary.
The primary will nack this if the startup_seqno is older than its one, and hence the joiner will fetch a more recent snapshot.

The most common form of this problem is if the primary was one of the original ones in a long-lived ledger, and the joiner has no snapshots.
In this case the comparison is 0 = 0 and so the joiner is allowed to join.

The primary should express an opinion about what snapshot is ok to join with.

Several options here:

• At least as modern as the most recent snapshot
• Given the snapshot-GC, the joiner should be at least as modern as max(min(snapshot), startup_seqno).

[cjen1-msft]

Proposed behaviour:
Add a flag to the join request 'retry_count'
On primary receiving a join request:

• if retry_count = 0: bounce if join_seqno < latest_snapshot
• otherwise: bounce if join_seqno < startup_seqno

This matches the desired behaviour of generally always using a recent snapshot, while preventing chasing and other bad behaviour.