video_mixer: vectorize host's video inputs

With the previous code, the video mixer only supports 2 inputs via
switchInput and switchSecondaryInputs. Both of those methods are
deprecated in favor of requestMediaChange.
Now, instead of two separate variables to manage, the video mixer
manages a vector of video inputs for the host. This will allow
the conference to have a better management of inputs via
requestMediaChange().

jami-project#1429

Change-Id: Ia65b35707bddf979154527c856e0574d96bbe519