Improve stability when massive DM-workers deployed on one node

[lance6716]

Feature Request

Is your feature request related to a problem? Please describe:

current architecture requires user deploy more than N DM-workers when migrate N sources, and many users tend to deploy multiple DM-worker at one node.

When that node is down/up, some etcd actions may be timeout. We need test about that.

And tiup-dm may not well support deploying many DM-worker on one node.

Describe the feature you'd like:

Test and improve it.

Describe alternatives you've considered:

Teachability, Documentation, Adoption, Migration Strategy:

[lance6716]

cc @lichunzhu