Use parallel make for builds. We use the GNU coreutils'
"nproc" utility to determine the number of cores available
for us to use. We then tell make to launch up to that number
of processes (-j$(nproc)) in parallel, but only if the system
load is below $(nproc) (-l$(nproc)).
That should make reasonably good use of however many cores github
makes available to us, while not overloading the build server
or the build server VM.
As 4316f75ed4 has removed gp2ddb
with its use of flex and bison, we do not need flex and bison
any more, and so the github workflows do not need to install
flex and bison either.
