Use provider datacenter as default when the agent cannot be reached

[remilapeyre]

According to the current documentation, the datacenter attribute should
default to the provider configuration and if it is not set to the
datacenter of the agent terraform is connected to:

* `datacenter` - (Optional) The datacenter to use. This overrides the datacenter in the
  provider setup and the agent's default datacenter.

This is not what the code actually does and the datacenter in the provider
configuration is not used most of the time.

Since changing the plugin behavior to what is actually documented may
break code relying on the current one. To keep backward compatibility,
we can change the documentation to:

* `datacenter` - (Optional) The datacenter to use. This overrides the
  agent's default datacenter and the datacenter in the provider setup.

and fall back to the provider configuration only when the datacenter of
the agent to which terraform is connected to can not be read (when ACL
is enabled and the token used by terraform does not have permission to
read /agent/self for example).

The change is very noisy as it requires to change the ConfigureFunc of
the provider from (*consulapi.Client, error) to (*Config, error) and
update all places where was being used.

The interesting change is at https://github.com/terraform-providers/terraform-provider-consul/pull/105/files#diff-7c162fe0d0220927a53a0f1977a47a71R361

Discussion: #57

[pearkes]

This looks good – a lot clearer and more accurate from my perspective. I tried to think of potential negative impact of this in terms of existing users configuring at each level but I assume that anyone going the route of per resource configuration won't notice any issues.