Turn off intent recognition from the execution model

[gdaniel]

This feature would be useful for the live chat: once the user has started a conversation with the support we don't want to bot to interfere in the conversation, and we want to forward all the user's messages to the support channel.

This could be done by disabling intent recognition and returning the default fallback for all the user's inputs. The recognition should be enabled again from a specific command/intent/event.

A possible way to do it is to set a boolean flag in the session, that would be checked before actually matching the user intent.

To discuss with @jcabot