fix: Queue batch commands to backlog during MOVED reconnection

[advaMosh]

Problem

When a MOVED-to-same-endpoint triggers reconnection (as implemented in PR #3003), batch commands (IBatch) issued during the ~50-100ms reconnection window fail immediately with RedisConnectionException: NoConnectionAvailable. This is because:

1. RedisBatch.Execute() calls SelectServer() which returns null for disconnected/reconnecting servers.
2. PhysicalBridge.TryEnqueue() returns false when !IsConnected, unlike TryWriteAsync/TryWriteSync which queue to the backlog via QueueOrFailMessage() → BacklogEnqueue().

This creates an inconsistency: individual async commands (e.g. db.StringSetAsync()) are queued and succeed after reconnection, but batch commands throw immediately during the same reconnection window.

Impact

• Applications using IBatch see 1-3 RedisConnectionException errors per MOVED redirect, while individual async commands see zero errors in the same scenario.
• This affects deployments where Redis is behind DNS records, load balancers, or proxies that use MOVED-to-same-endpoint.

Fix

RedisBatch.Execute(): When SelectServer() returns null and BacklogPolicy.QueueWhileDisconnected is enabled, retry with allowDisconnected: true to locate the server endpoint. This mirrors the pattern already used in PrepareToPushMessageToBridge().

PhysicalBridge.TryEnqueue(): When !IsConnected || NeedsReconnect and BacklogPolicy.QueueWhileDisconnected is enabled, queue messages to the backlog instead of returning false. This matches the existing behavior of TryWriteSync() and TryWriteAsync() for individual commands.

Testing

Tested with a proxy that sends MOVED-to-same-endpoint followed by connection close:

• Before fix: 1-3 NoConnectionAvailable errors per redirect event (batch.Execute throws during ~50-100ms reconnection window)
• After fix: 0 errors — batch commands are queued to backlog and sent after reconnection completes

The fix only activates when BacklogPolicy.QueueWhileDisconnected is true (the default), preserving existing behavior for users who have disabled backlog queuing.

Related

• Handle MOVED error pointing to same endpoint. #3003: Handle MOVED error pointing to same endpoint (the original MOVED reconnect fix)
• MOVED redirects to the same endpoint are not retried #2990: Original issue report for MOVED-to-same-endpoint

[advaMosh]

hey @mgravell , thank you for the quick response !
I see the failing tests in the failing check are flaky regardless to this PR. can / will it be merged in that state?
also - Do you have an estimate for when this PR is expected to be merged, and which SE.R versions it’s likely to be included in?

[mgravell]

The CI has a bit of noise, sadly. I've poked it. I'll do another eyeball tomorrow, but there's no reason to delay this any - we tend to build fairly regularly when there's reason to.