lloidparts: remove queue between firbank and matrixmixer

- a FIXME says to do so when a problem in the collect pads class has been
  addressed.  that problem has been addressed, but we have also altered our
  use of the GstBaseTransform class so that every input buffer always
  produces an output buffer (just, sometimes, with 0 size), which removes
  the need to have fixed the collect pads class
- this causes the firbank and matrixmixer to run in the same thread when
  the otho-gate-fap feature is turned off, but we generally have so many
  more of these elements than there are CPU cores there's no way the
  additional parallelism could have been used anyway.