TiDB should retry on every PD client URL until success or failure

[JmPotato]

Bug Report

1. Minimal reproduce step (Required)

1. Start a cluster with 3 PD instances.
2. Kill the PD server that is first in the list of members returned by the pd.GetMemebers and do not reboot it.
3. Try show config SQL on TiDB.

2. What did you expect to see? (Required)

TiDB executes the SQL successfully.

3. What did you see instead (Required)

Root case

GetPDServerInfo should retry on every PD client URL until success or failure. In this case, if the first member is the PD server that crashes, it will never work until the PD server reboots.

tidb/infoschema/tables.go

Lines 1658 to 1669 in 7500d79

	for _, addr := range members {
	// Get PD version, git_hash
	url := fmt.Sprintf("%s://%s%s", util.InternalHTTPSchema(), addr, pdapi.Status)
	req, err := http.NewRequest(http.MethodGet, url, nil)
	if err != nil {
	return nil, errors.Trace(err)
	}
	req.Header.Add("PD-Allow-follower-handle", "true")
	resp, err := util.InternalHTTPClient().Do(req)
	if err != nil {
	return nil, errors.Trace(err)
	}

[JmPotato]

/label affects-6.1, affects-6.0, affects-5.4, affects-5.3, affects-5.2, affects-5.1, affects-5.0, affects-4.0

[jebter]

/label affects-6.1
/label affects-6.0
/label affects-5.4
/label affects-5.3
/label affects-5.2
/label affects-5.1
/label affects-5.0
/label affects-4.0