Clean up NoWaitForReady

[jsha]

In #6834 we added a config NoWaitForReady that allows gRPC requests to instantly error when there are no backends available (vs waiting for a backend and eventually reporting a timeout). That config is now enabled in prod. We should deprecate and ignore it, effectively never waiting for readiness. As part of that process we should also search the codebase for FailFast (which is the older gRPC terminology for the feature) and clean up relevant comments.