So I played with the Shark audio a bit.  I found 4 problems, the first
two of which I've fixed:

1) The play parameters were being used on the record channel.

2) ess_drain() was hanging forever.

3) There is an interesting hardware issue.  Both the play and record
channels must run at >=22050Hz, or both <22050Hz; otherwise the wrong
crystal is used on the play channel, and it plays at half or double
speed.  (This was masked in some cases by #1.)

4) At 44KHz, I get `warbling' output, which sounds very similar to the
output of the SB driver before it did auto-initialize DMA.