Replace health checking with server-capability getter and disable retries on internal errors

[cretz]

What was changed

• Remove health checking
• On every connection, call WorkflowService.GetServerInfo to eagerly obtain the capabilities of the server (or ignore if call is not there, it was added in New GetSystemInfo RPC temporal#2309)
• If the capabilities say we don't need to retry internal errors, don't

Why?

We need to know from the server what it supports. Specifically in this case, we need to stop retrying internal errors if the server says it differentiates between them.

Checklist

1. Closes Implement server capabilities and replace health check #701

⚠️ WARNING - Breaking API Change

This makes the following backwards incompatible changes:

• Removes client.ConnectionOptions.DisableHealthCheck, client.ConnectionOptions.HealthCheckAttemptTimeout, and client.ConnectionOptions.HealthCheckTimeout. This means users of those options will have to change code on upgrade.
  • Since on-new-client health checking (default performed, but opt-out via DisableHealthCheck) is replaced by obtaining these capabilities, we removed the configuration options.
  • We now offer no option to disable eager connection to the server. This was chosen as the better option to allowing callers to disable our ability to obtain server capabilities.
    • This will require reworking for anyone using DisableHealthCheck (which includes our own server), to not create the client until the server is running.
    • Arguably one should not use a client that cannot connect to a server, and connecting later instead of fail-fast makes troubleshooting tougher.
  • It was decided to remove the options instead of deprecate them since they wouldn't be deprecated, but rather no longer used.
• Removes health checking from NewNamespaceClient.

The removal of health checking and the removal of eager-connect opt-out was discussed internally with @bergundy, @alexshtin, and @yiminc. We can of course take other approaches if there are concerns here.

@mfateev and @samarabbas - Please review.

[cretz]

Note, I am having CI failures I am struggling to replicate locally. Pardon the repeated forced pushes.

[cretz]

After discussion here, we can't reasonably require a namespace to be ready for cross-namespace clients. We are going to have to remove namespace from this call probably. I will update with more info after discussion.

[cretz]

I will be waiting for after next server release and after next SDK release before coming back to this

[cretz]

This has now been updated to remove namespace from the call and is ready for review/inclusion.

[bergundy]

I'm fine with not exposing GetSystemInfoTimeout here until it's requested.