This behavior was requested by my one user on NetBSD, and after checking
with Stefan Esser of FreeBSD, we decided that it wasn't a big enough
deal to not do. In fact, in my mind, it fixes a lot of problems with how
B/DC_ENV_ARGS interacts with scripts.

This small non-portable thing was doing something I thought was
portable: setting IFS=$'\n'. Brian Callahan of OpenBSD ran into a
problem though, and after I found and fixed the offending bit, I checked
the standard. It is not in the standard for POSIX sh. Sigh...

I can remove this without problem, by the way. That's because it was in
there for safety only, safety for if I used source file names with
spaces, something I haven't done, nor was I planning to. I was just
coding carefully. But since setting IFS is not portable, I have to stick
with file names with no spaces. That's fine, and if I forget, the build
should loudly break.

This small annoyance made the test output slightly different from what
it was before I changed the build system. I changed it back.

The gcc on OpenBSD complains with the combined #if in status.h of
defined(__has_attribute) && __has_attribute(fallthrough), so I split
them.

They are under a new test target: run_all_tests.

This is what happens when you use broken shells on Linux. This was found
by Stefan Esser of FreeBSD, probably because the shells on FreeBSD are
more sane.

This is for when I want to do a reconfigure, like compiling with ASan
for checking all of the fuzzer output, but I still need the fuzzer
output.

This bug was allowing fatal errors to slip through and trip an assert
that I put in for that purpose. I just made sure they don't slip past
anymore.

Scripts take the longest, so they should be started as soon as possible.

This may not seem important, but when I run the release script, it will
matter, especially on the ones that take a long time because a parallel
`make test` is about a third as long.

I may back out this change if it doesn't work on non-GNU makes.

This does this because FreeBSD does not have /dev/urandom, but they do
have /dev/random.

This commit adds a fuzz build option (-z) and also tells afl where to
fork() for best efficiency. I tried persistent mode, but there was a bit
of state that I could not reset properly, so I just called it there.

This one was exactly like the other one, except in ocurred in the
*other* use of bc_program_assignStr(). In both cases, I did not
correctly differentiate between the two cases for strings, whether in
the result or in the number (both are possible).

After finding this crash (afl++ found it), I went through all occurences
of strings to see if there were any others, and I could not find any.

I have had this infrastructure outside of the tree ever since I started
fuzzing bc. I decided that, at the same time I rebuilt my computer to
have 16 cores instead of 8, I would redo the infrastructure to make it
usable for posterity that would like to maintain bc after I get hit by a
bus.