unify/refactor dynamic timeslot code
The two kinds of dynamic timeslots could use the same data structures.
The IPAC style TCH/F_PDCH was implemented first, using ts->flags.
The Osmo style TCH/F_TCH/H_PDCH needed more detailed state, so ts->dyn was added.
The TCH/F_PDCH could also use ts->dyn.
Also, we had neverending bugs of places that forgot to check for dyn ts.
The code would be less fragile if we kept the actual current mode the
timeslot is in in the actual ts->pchan field, where all the code expects it.
Currently we keep "TCH_F_TCH_H_PDCH" in pchan and store the current real
mode in dyn->pchan_is. So all code needs to be aware of dynamic channels
and has to add checks to find out the current mode.
I assume it would be nicer if the fact that a channel is dynamic is kept
in ts->dyn, and the ts->pchan can be evaluated like with static channels.
ts->pchan should probably also reflect switchover as an invalid pchan.
This way all code can just directly use ts->pchan, and the places that
are concerned with switchover can look it up in ts->dyn.*.