This resolves an issue where multiple versions of the symbols are resolved due to them living in the two DSOs.
Fixes issue #12.

ensures that frame time will reflect any
discontinuity in processing (xrun or skipped cycles). It also cleans up
the code related to maintaining the DLL and frame time a bit, all of it
is now together in jack_run_cycle() and the logic behind it is easy to
follow.

The 'delayed_usecs' argument to jack_run_cycle() is now probably
redundant, the value computed locally (the 'dus' variable) is in
all cases more accurate than what the backend can provide.

Without cycle counting, CPU scaling isn't necessarily bad anymore, so we
don't have to worry the user.

The purpose of this commit is to get rid of JACK_TIMER_CYCLE_COUNTER.

The cycle counting clock is no longer supported. For backwards
compatibility with scripts, allow the user to request the cycle clock on
the command line, but use the system clock instead.

Note that the -I option is still the correct thing to use if you want more control over setting up one of these clients.

When reading the byte from the wait file descriptor failed, jack would mark the client at the head of the external client chain
with an error. But the fault may not have been with that client. So now if the read error is EAGAIN ('no data available') do
not mark the client, but rely on other mechanisms to detect too-slow or dead clients.

Also tweak VERBOSE output a little in some related spots.

we used to still look in ADDON_DIR, which might be "secure" but is a bit pointless and inconvenient

* move port UUID initialization into engine so that it is actually shared across clients
* export jack_server_dir() within libjack