Made some changes to async buffer, to better accommodate output dma

Made some changes to async buffer, to better accommodate output dma
(added read allocation).  Tweaked memory barrier usage for async buffer
to try and improve its correctness.  comedi_buf_munge() is now handled
entirely inside comedi_buf_write_free().  Added some helper functions
mite_sync_input_dma() and mite_sync_output_dma() for use in
driver interrupt handlers.  Added beginnings of support for NI general
purpose counter input commands.