Can not create changefeed when start ts is less than current ts - gcttl

[overvenus]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. What did you do? If possible, provide a recipe for reproducing the error.

Create a changefeed with a start ts that is close to gc safepoint.

# Change gc lift time to 2400h
mysql -h 172.16.5.33 -P 47904 -u root -E -e 'update mysql.tidb set VARIABLE_VALUE="2400h" where VARIABLE_NAME="tikv_gc_life_time";'

# Wait 24 hours (the default gc-ttl)

tiup cdc:nightly cli --pd="http://172.16.5.33:47902" changefeed create --sink-uri="blackhole://" -c "bl2" \
  --start-ts=$(( $(pd-ctl -u http://0.0.0.0:47902 service-gc-safepoint | jq '.service_gc_safe_points[] | select(.service_id == "gc_worker") | .safe_point') + 1000 )) --no-confirm --sort-engine=unified

tiup cdc:v5.0.0-nightly cli --pd="http://172.16.5.33:47902" changefeed list
The component `cdc` version v5.0.0-nightly is not installed; downloading from repository.
component cdc version v5.0.0-nightly is already installed
Starting component `cdc`: /home/stn/.tiup/components/cdc/v5.0.0-nightly/cdc cli --pd=http://172.16.5.33:47902 changefeed list
[
  {
    "id": "bl1",
    "summary": {
      "state": "normal",
      "tso": 425199630964752529,
      "checkpoint": "2021-05-26 13:44:29.586",
      "error": null
    }
  },
  {
    "id": "bl2",
    "summary": {
      "state": "stopped",
      "tso": 425043878316868600,
      "checkpoint": "2021-05-19 16:42:00.386",
      "error": {
        "addr": "172.16.5.33:47913",
        "code": "CDC:ErrSnapshotLostByGC",
        "message": "[CDC:ErrSnapshotLostByGC]fail to create or maintain changefeed due to snapshot loss caused by GC. checkpoint-ts 425043878316868600 is earlier than GC safepoint at 425177025946845484"
      }
    }
  }
]

pd-ctl tso 425043878316868600
system:  2021-05-19 16:42:00.386 +0800 CST
logic:  1016

pd-ctl tso 425177025946845484
system:  2021-05-25 13:47:18.286 +0800 CST
logic:  300

2. What did you expect to see?

Create a changefeed successfully.

3. What did you see instead?

Failed.

4. Versions of the cluster

Latest release-4.0.