- I need to migrate and/or rewrite the swamp 1.1.0 documentation to the
  sslswamp man-page. Right now the man page is very incomplete.

- Need to fix up snoop for the new libnal model.

- All the executables need a review to make sure they're safe w.r.t. proper
  use of NAL_LISTENER_finished() and also NAL_SELECTOR_new_objects() for
  program termination.

- snoop does not do any flush-handling when one side of the proxy disconnects,
  but immediately closes the other side and destroys the context. What should
  happen is that we should not close the other side until all data accepted
  from the closing side has been sent out the other side.

- I'm pretty sure the building of the documents is not 100% kosher. Eg. the
  links don't seem to do anything useful in the HTML versions at least, and
  there's loads of cross-link errors/warnings when building. (NOTE: I've
  disabled compilation of HTML documents for now).

- dummy FIFO connections created using NAL_CONNECTION_create_dummy should be
  trivially selectable if anyone's daft enough to try. Ie. if you add a FIFO
  to a selector, the select() should break instantly if the FIFO is non-empty
  otherwise the FIFO should have no influence at all on the real select(2).
  Right now, NAL_SELECTOR_add_conn() silently ignores dummy connections
  completely.

- dc_test should in theory avoid the case that a session it added could have
  been expired (we can't say definitively because of races). If it gets within
  a couple of seconds of expected session expiry, the session should be marked
  unusable by dc_test for further cache requests until at least a couple of
  seconds after the expiry where dc_test should then assume it was removed by
  the server so future ADDs succeed, DELETEs fail, etc.

- Since overhauling libnal for the alternative selector model, piper ceases
  working between dc_test and dc_server. It works between nal_ping and nal_echo
  though, so it's probably a layer in libdistcache or something (sigh).