Currently the only way that the UI can know whether connect is enabled
or not is whether we get 500 errors from certain endpoints.

One of these endpoints we already use, so aswell as recovering from a
500 error, we also remember that connect is disabled for the rest of the
page 'session' (so until the page is refreshed), and make no further
http requests to the endpoint.

This means that log spam is reduced to only 1 log per page refresh
instead of 1 log per service navigation.

Longer term we'll need some way to dynamically discover whether connect
is enabled per datacenter without relying on something that will add
error logs to consul.