There's something broken in the "profile enabled" logic at the moment which
makes server_is_enabled return FALSE sometimes when it shouldn't which causes
the necessary interface not to be added to remote device objects. Use simply
the config file values for now since we don't support per-adapter server
configuration anyway.

Previously the avctp session would be free'd and the connect callback would try
to access this free'd memory. Now we detect this situation properly but the
fundamental problem remains that the bt_l2cap_connect API doesn't allow
aborting the connect procedure.

Remove the stream state callback if stream start fails. Without this patch the
callback could still be called after a unix client has disconnected which in
turn would cause a double-free to occur. There is code to remove the callback
also in client_free() but it won't work since we remove any references to the
session and stream in the resume callback.

With this patch drivers can notify the core daemon that a device supports some
UUID. This is useful in case we don't have a service record for a profile which
the remote device initiates a connection for.

This is preparation for allowing the device object to be correctly set up even
when we don't have the full service records but only the remote UUID's.