Fix race condition allowing duplicate unbuffered reads

[dmac]

Previously, this sequence of events would result in the same
value being read twice from an unbuffered channel:

Thread S                     Thread R
--------                     --------
unbuffered_chan_send()
  chan->data = data
  chan->w_waiting++
  cond_wait on chan->m_mu
                             unbuffered_chan_recv()
                               mutex_lock chan->m_mu
                               chan->w_waiting > 0
                               *data = chan->data
                               cond_signal chan->w_cond
                               mutex_unlock chan->m_mu

                             unbuffered_chan_recv()
                               mutex_lock chan->m_mu
                               chan->w_waiting > 0
                               *data = chan->data
                               cond_signal chan->w_cond
                               mutex_unlock chan->m_mu
  chan->w_waiting--
  mutex_unlock chan->m_mu

Moving the chan->w_waiting-- to immediately after the data
is read in unbuffered_chan_recv() prevents this double read.

Fixes #23

[tylertreat]

Thanks!