statement-store: reject statement peers during major sync

[peter941221]

Summary

Fixes #11411 by preventing statement peers from remaining half-connected during major sync.

• reject inbound statement substreams while major syncing
• disconnect streams that still open during that window
• disconnect already-connected peers that send during that window
• keep reconnect + initial sync as the recovery path
• keep initial-sync telemetry consistent on these disconnect-driven cleanup paths

This also includes the stream-open snapshot fix so a peer admitted just before major sync starts still queues initial sync correctly instead of falling into the same broken state through a race.

Why this shape

Silently ignoring statements keeps the broken middle state alive:

• the statement stream looks open,
• live statements are not accepted,
• and there is no clean replay trigger later.

Disconnecting is the smallest safe repair because it reuses the existing lifecycle instead of adding new sync-transition machinery.

Tests

• rejecting inbound statement substreams while major syncing
• disconnecting streams that open while major syncing
• disconnecting already-connected peers that send during major sync
• reconnecting after sync exit and re-queuing initial sync
• preserving initial sync if major sync starts mid stream-open
• not recording major-sync disconnect cleanup as a completed initial sync
• decrementing initial_sync_peers_active when flooding cleanup removes pending sync state

Validation

• git diff --check refs/remotes/upstream/master..HEAD -- substrate/client/network/statement/src/lib.rs
• cargo test -p sc-network-statement --lib major_sync -- --nocapture

Note: the full lib suite on this narrowed branch still hits the older sustained-flooding test that 7bfd50e later stabilizes, so that test-only stabilization can remain a separate follow-up if preferred.

[cla-bot-2021]

User @peter941221, please sign the CLA here.

[peter941221]

Quick note on the follow-up commit:

• the third commit does not widen the behavioral scope beyond the core major-sync fix
• it only keeps initial-sync telemetry consistent on disconnect-driven cleanup paths
• specifically, it avoids counting aborted initial sync as completed, and keeps initial_sync_peers_active balanced when pending sync state is removed during disconnect cleanup

So the main slice is still the same narrow one:

• reject / disconnect statement peers during major sync
• preserve reconnect + initial sync as the recovery path
• keep the stream-open transition race closed