Kx,Ky values were just plain wrong. Instead of the
default value of 10_000_000 it was 1_000_000. Ksum
was using the value 1.0 with old value FIX25_24
that was used by the previous DSP chain. Now,
we use FIX25_0.

As we now have, in the same register, two fields,
acha_chan and acq_dtrig_chan, we must modify only the
bits associated with that field. Previously, when
we set the acquisition channel, we were rewriting the
whole register, effectively selection channel 0 for
the acq_dtrig_chan field.

Now, device names are created as
/dev/fpga-<bus_number << 8><device_fn> and as
/dev/fpga/<bus_number << 8><device_fn>

This improves portability, as zsys_daemonize ()
is system independent.

As we do not send the $TERM message anymore,
it makes sense to rename the function to a more
suitable name.

On DEVIO destruction, typically triggered by the
SIGINT, the main DEVIO thread receives it and
used to start sending $TERM messages to all SMIOs
(implemented as actors/threads).

However, the SMIOs also receive the signal and exited
gracefully, as usual. The problem arises by the fact
that some SMIOs exited before the DEVIO had a chance
to send the message to that specific SMIO. This would
block the zmsg_send (), as it waits forever to send
the message by default, which will never get to the
SMIO, as it does not exists anymore.

The solution would be to have a timeout for the
zmsg_send () and consider the SMIO as nonexisting
if it cannot send the message some time after.

Instead of doing this, we simply omit the message send
and call zactor_destroy () directly, as it already
implements this timeout.

The LGPLv3 license is not suitable to our project
as it is not a library, for instance. GPLv3 is still
compatible with our libraries (LGPLv3 for libzmq)
and provides us with roughly the same benefits.

Travis cannot run ldconfig on install, so we
tell our build system not to execute it.

This is important on systems that don't have
ldconfig or on test systems, such as travis

As we use -Werror all warnings are treated as
errors and liberrhand will issue an warning if
we do not specify DEBUG flags. To keep things
clean, we ignore these warnings here.

We were not destroying the zrex structure as
we exit prior to that on success.

Hashes (zhash or zhashx) take a copy of the
key value and so they must be free locally too.

We "have" (easier as the functionality required is
already implemented) to use the LLIO library to
read/write in our PCIe devices.

We already build libbpmclient with LIBS
command. No need to run it twice.

This happens with liberrhand using #warning
to inform the user that we didn't set the debug
level and it will use the defaults.

Now, we must build and install all of the
libraries prior to the main application.
This can be done independently for each
library or using the new libs_compile_install
target.

As LLIO is build as a library, we use
the new header file location

Now LLIO is build as a library and we must
not use direct references to it.