[vSphere Provider] Fix the bug that multiple worker types doesn't work

[JingChen23]

Description

Currently our code assumes that there is only one worker node type.
In this change I fix the bug to let it support multiple worker node types.

Test

Added UT assert to cover the new case.

Used a yaml snippet like this:

available_node_types:
    ray.head.default:
        resources: {"CPU": 8, "Memory": 8192}
        node_config:
            resource_pool: test
            datastore: vsanDatastore
    worker:
        min_workers: 1
        max_workers: 5
        resources: {"CPU": 4, "Memory": 8192}
        node_config:
            resource_pool: test1
            datastore: vsanDatastore
    worker1:
        min_workers: 1
        max_workers: 5
        resources: {"CPU": 2, "Memory": 4096}
        node_config:
            resource_pool: test1
            datastore: vsanDatastore

Verified that the workers are in the expected resource pool, the 2 different node types resources are expected:

Also the Ray cluster looks good from the dashboard:

[architkulkarni]

We don't need datastore anymore?

[JingChen23]

Please forgive the original code which is hard to read.

Before the change, the logic is: if the datastore of the worker node is unset, use the head node's datastore by default.

But, what we actually want is (which is also published in our documentations): if the datastore of the worker node is unset, use the frozen VM's datastore.

So I make this change. We left the datastore empty for the worker node. It will automatically use the datastore of the frozen VM with this empty parameter.

[architkulkarni]

Looks good, one minor question

[architkulkarni]

Linkcheck and chaos test and HA tests are unrelated, this PR only touches the vSphere cluster launcher.