lightning: tidb restart cause lightning to cost more time to import

## Bug Report

Please answer these questions before submitting your issue. Thanks!

### 1. Minimal reproduce step (Required)

- 3 TiDB pods. killing one of them every 10 mins.
- The command per second keep going down when the chaos testing starts and never recover until turning off the chaos testing.
- The import time increase from 1h30m (no kill) to 2h41m (kill)
- The job finished successfully.
- From the log there are some execute statement failed but looks they success after retry.
- From grafana, tidb-0 restart 1 time, tidb-1 restart 2 times and tidb-3 restart 3 times. No obvious gap between each restart
- No tikv restart
- data file size is 63G, importing an existing table which already has 64G data

### 2. What did you expect to see? (Required)
There is about 10 mins delay due to TiDB restart.

### 3. What did you see instead (Required)

The actual delay was 1h11m.

The time some file complete restore:
```
[2022/03/13 17:51:42.867 +00:00] [INFO] [restore.go:2450] ["restore file completed"] [table=`test`.`?`] [engineNumber=1] [fileIndex=1] [path=test.?.00482.parquet:0] [readDur=4.099086015s] [encodeDur=1.936240902s] [deliverDur=2m50.256444304s] [checksum="{cksum=0,size=418166857,kvs=955926}"] [takeTime=2m50.46770005s] []+
```
Some file took a lot of time.
```
[2022/03/13 19:02:36.839 +00:00] [INFO] [restore.go:2450] ["restore file completed"] [table=`test`.`?`] [engineNumber=1] [fileIndex=33] [path=test.?.00546.parquet:0] [readDur=4.448735371s] [encodeDur=2.02285467s] [deliverDur=1h2m49.443847954s] [checksum="{cksum=0,size=429542501,kvs=1033445}"] [takeTime=1h2m49.669141098s] []_
```

Some "driver: bad connection" errors occurred during TiDB restarting. Most errors were [error="invalid connection"].

### 4. What is your TiDB-Lightning version? (Required)

v5.3.0