[CI] Bump tvmai/ci-cpu to 0.63 in CI

[comaniac]

After #5936, the latest docker image with DNNL is tvmai/ci-cpu:0.63.

cc @tqchen@zhiics

[tqchen]

We cannot yet directly merge it(merging will result in ci outrage), because of the CI error mentioned (http://ci.tvm.ai:8080/job/temp-ci-docker-staging/job/ci-stage/39/execution/node/324/log/). It won't be not reflected in the current PR CI because the change won't take in effect before it is merged(still running the old image)

[comaniac]

Ah...then would you suggest closing this PR first and you will file another one after the issue has been resolved, or we just leave it here for now?

[tqchen]

Let us close this for now, I will file another one when the issue get resolved, sorry for the confusion. We will likely need to update tensorflow to 2.2 to resolve the error

[comaniac]

Sure no problem. Will also leave #5919 open during this time.