owner: fix memory accumulated when owner consume etcd update slow

[amyangfei]

What problem does this PR solve?

Fix for a memory accumulated issue in owner

https://github.com/pingcap/ticdc/blob/eed57c94491fd140f0a8a0528acc574bc542bb0b/cdc/owner.go#L1024-L1039

The output channel consume speed (which relays on how often L1035 can be run) may be much slower than the etcd key update frequency, which will lead to data accumulated in etcd watch client

What is changed and how it works?

Ignore position update event if output channel is full, it is safe enough as the owner has a builtin ticker to trigger o.run

NOTE: this is a release-4.0 quick fix, I will update master/release-5.0 and add test cases later

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below) (Have tested in an internal test cluster)
  • setup a ticdc cluster, create one or more changefeeds
  • running cdc for 10mins, get memory profile of cdc owner by http://<ip>:<port>/debug/pprof/heap
  • find the memory usage of etcd.clientv3, should be less than 5MB

Release note

• Fix a bug that too much memory may be consumed in etcd watch client in cdc owner.

[amyangfei]

/run-all-tests

[zier-one]

LGTM

[amyangfei]

/run-all-tests

[amyangfei]

/run-all-tests

[liuzix]

LGTM