Fix parallel builds and use them in gh workflow

[ndim]

The old gettext 0.18.3 po/Makefile.in.in has problems when used with parallel make (make -jN with N>1). Running make -j$(nproc) -l$(nproc) distcheck on a 12 core system fails reproducibly.

gettext 0.19.1 has apparently fixed those problems with parallel builds, as running make -j$(nproc) -l$(nproc) distcheck on the same 12 core system reproducibly succeeds with gettext 0.19.1's po/Makefile.in.in file

When parallel builds work, we can now also make use of them during the CI check github Action workflow, so we change all make commands there to make -j$(nproc) -l$(nproc).

This is a PR to trigger and test the CI builds, and to allow discussion about requiring gettext 0.19.1 which is a release not even 8 years old, making it the youngest build tool release required to build libgphoto2.

[ndim]

While we do not have proper benchmarks from the Github Actions, we can see that adding -j2 -l2 to the make command lines consistently decreases times, sometimes significantly. As the -j2 checks have run

step	-j1	-j2	same	sleep iff isatty(1)	same
SHA	82251c0	eb63ef9	eb63ef9	d26577c	d26577c
CI LOGS	725	727	731	728	730
autoreconf	24	20	24	18	26
configure	20	16	20	15	21
make all	61	27	35	27	34
make check	4	2	2	1	3
make install	5	5	5	4	5
make installcheck	n/a	n/a	n/a	n/a	n/a
make distcheck	137	87	104	83	104
TOTAL	278	180	219	176	225

The isatty column produces about the same values as the -j2 column as it only affects builds with non-standard camlib set. However, the -j2 column and the isatty column appear to have run on a faster runner than the -j1 and same column. The same column is the same code as the isatty column, but run again, apparently on a similar runner as the -j1 column. So the best columns to compare for evaluating this PR's effect are the -j1 column vs the same column.

The few seconds here and there are not significant, but reducing make all to about 55%, reducing make distcheck to about 75%, and the total runtime to about 80% is a lot better than nothing, keeping in mind we are only running with an nproc value of just 2 on our Github Action runners.

[ndim]

step	-j1	-j2	sleep iff isatty(1)	--with-camlibs=everything
SHA	82251c0	eb63ef9	d26577c	36a3666
CI LOGS	725	727	728	729
autoreconf	24	20	18	19
configure	20	16	15	16
make all	61	27	27	42
make check	4	2	1	1
make install	5	5	4	12
make installcheck	n/a	n/a	n/a	n/a
make distcheck	137	87	83	118
TOTAL	278	180	176	235

FWIW, on the faster runners (autoreconf about 20 seconds; the slower runners take about 25 seconds for autoreconf) with -j2 we could even build with --with-camlibs=everything and still be faster than we were with -j1 and the default camlib set. Within one set of runners, we would probably be about even, or still a bit slower.

I am not advocating for or against adding --with-camlibs=everything to the CI workflow here. If anything, I would add DISTCHECK_CONFIGURE_FLAGS=--with-camlibs=everything to the make distcheck command. But this is probably also not worth the CPU cycles.

[msmeissn]

change looks good to me.