decodebin3: Make input activation "atomic"

decodebin3: Make input activation "atomic"

When adding inputs dynamically, we need to make sure the new parsebin are
added *and* activated by the same thread (by taking the state change lock).

The rationale for this is that the calling thread might be an upstream streaming
thread and when activating parsebin it might call back upstream. If we don't use
the same thread (ex: when the application does a state change on decodebin3
between the moment we add parsebin to decodebin3 and we synchronize the state of
parsebin) then we would end up in different threads trying to take upstream
recursive locks.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/merge_requests/932>