Thanks Ilya Khaprov for the fix.

On FreeBSD, libsodium's `sodium.h` header is installed in
`/usr/local/local`. The Makefile already adds
`/usr/local/include/sodium` to the compiler's `-I` search path, but it
doesn't cover the FreeBSD case.

This patch adds `-I /usr/local/include` to the CFLAGS on the make(1)
command line to let libsalty2's Makefile pick it up. This fixes the
build on FreeBSD.

Its Makefile called sed(1) in a way that was only compatible with GNU
sed. As a consequence, the build failed with BSD sed.

Now, the test case fetches dependencies and patches `reloader`'s
Makefile to use a sed(1) script compatible with both implementations of
sed(1).

This is the standard name and is available everywhere, unlike "netcat".

If `V=...` is set on the command line to increase verbosity while
debugging, the first grep(1) command will also catch the debug output of
make(1) which contains the executed command line and thus mentions
"hello".

The fixed grep(1) command line explicitly grep for a line starting with
"hello".

... instead of https://github.com/hypernumbers/hypernumbers which is no
longer available apparently.

At least for me with Erlang/OTP 28, parse_transform messages end up on
stderr.

Let's always redirect stderr to stdout. It will work for anyone with any
versions of Erlang.

If the user running the tests has set `init.defaultBranch` to something
else in their `.gitconfig`, the tests won't have the expected branch
name.

This patch explicitly sets the branch name to make sure the tests are
not affected by a user setting.