refine gc strategy of pump (#646)

[IANTHEREAL]

What problem does this PR solve?

most users forget to mark drainer offline using binlogctl after they stop drainer, It would block gc of pump, and then the pump will easily fill up the disk capacity because there are too many binlogs.

Regardless of ease of use or design, this design is unreasonable. There are two main reasons:

• when the disk space of the pump is full, the pump will be unavailable, which will have a great impact on the data and TiDB instance.
• drainer is a relatively separate tool and should not have too much impact on core system formation

What is changed and how it works?

• pump only keeps binlog events whose ts <= current time - gc(days)
• issue a critical metrics while some drainer checkpoint is old than gc ts of pump，it means binlog that drainer need is purged

Check List

Tests

• Unit test

Related changes

• Need to cherry-pick to the release branch
• Need to update the documentation
• Need to update the tidb-ansible repository
• Need to be included in the release note

[IANTHEREAL]

/run-all-tests

[IANTHEREAL]

/run-all-tests

[suzaku]

/run-integration-tests

[mahjonp]

/run-all-tests tidb=release-2.1 tikv=release-2.1 pd=release-2.1

[suzaku]

/run-all-tests tidb=release-2.1 tikv=release-2.1 pd=release-2.1

[suzaku]

LGTM

[IANTHEREAL]

@july2993 PTAL

[july2993]

LGTM