add workqueue metric

[qiankunli]

tf-operator version :v1.2.1

我们发现一些tfjob 从创建到开始调度耗时很久，达到8分钟，初步结论是tf-operator cpu 配少了 导致消费workqueue 速度不够（tfjob 每天400+，pod 3800+），后来增加了cpu和 threadiness，现在看大致正常了。

Sep 16, 2021 @ 17:28:51.301 {"filename":"common/job.go:144","level":"info","msg":"Reconciling for job v1-tensorflow-0916171919266","time":"2021-09-16T09:28:51Z"}
 Sep 16, 2021 @ 17:19:29.115 {"filename":"tensorflow/job.go:118","job":"xdl-system.v1-tensorflow-0916171919266","level":"info","msg":"TFJob v1-tensorflow-0916171919266 is created.","time":"2021-09-16T09:19:29Z","uid":"e8b0d2c4-f876-4c54-b1e5-380fc4b4f92f"}

因此建议 将client-go workqueue 的metric加入到 tf-operator 的metric，这样分析延迟原因和证明问题解决 都方便一些，也便于用户作为调整cpu和 threadiness 的依据。

[gaocegege]

SGTM.

/kind feature

We should add metrics for the workqueue

[gaocegege]

/help-wanted

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.