WaitForState functions should be a bit more tolerant of http errors.

[Smithx10]

Currently during a provision if the client encounters an http issue or returns an error we return err and won't update our state. I discovered this while debugging an issue with AddNIC ( TritonDataCenter/sdc-cloudapi#32 )

I believe the correct behaviour here is to not error out and let the TTL expire if we never reach our desired state. I believe this is currently 10 minutes.

	stateConf := &resource.StateChangeConf{
		Pending: []string{oldName},
		Target:  []string{newName},
		Refresh: func() (interface{}, string, error) {
			inst, err := c.Instances().Get(context.Background(), &compute.GetInstanceInput{
				ID: d.Id(),
			})
			if err != nil {
				return nil, "", err
			}

			return inst, inst.Name, nil
		},
		Timeout:    machineStateChangeTimeout,
		MinTimeout: 3 * time.Second,
	}
	_, err = stateConf.WaitForState()

https://github.com/terraform-providers/terraform-provider-triton/blob/master/triton/resource_machine.go#L556