media: dvb: pipe work queue optimization

Having a pipe work object for each sps callback quickly
caused queue to fill-up when playing from memory where TSPPv2
hardware processing is very fast.
Each pipe now has its own dedicated queue, and in case the same
event repeats (which is the likely scenario) the events are merged
into one work object with event count instead.

Change-Id: I00b99f3d8de3718c73a238520d49e5bdb3843545
Signed-off-by: Gilad Broner <gbroner@codeaurora.org>