stuck check can still make some mistake reports

[hicqu]

What did you do?

Start one TiCDC instance, and run a changefeed with 30 tables. Then

1. pause the changefeed, and prepare some data in the upstream cluster;
2. inject a latency into the sink;
3. resume the changefeed;
4. after a while, we can get the log "stuck too long".

Why "stuck too long" happens?

It's because the table is the last one to finish incremental scan. There are already 29 tables scheduled into sink. After the last table gets scheduled, its resolvedTs has been updated, but its checkpoint can only be advanced after all other tables
pending data has been consumed. So, the stuck check is triggered, which is unexpected.

What did you expect to see?

No response

What did you see instead?

As decribed above.

Versions of the cluster

Master branch.