FreeBSD and MacOS variants of `sed` do not accept `-i` parameter without any argument.

Resolves: #354

Previously we only warned about these things in maintainer mode, and not
all of the warnings were set to error mode, which led to some issues
being missed.
By unconditionally applying some Werror flags, we will catch more issues
(style or actual bugs) faster, while the flags should also be safe to
use in production and shouldn't make the code fail to build with newer
versions.

Apparently, bootstrapping a system with Rust in the toolchain is still a
major pain, and since rsvg is being rewritten in Rust, AppStream is
dragging it in if compose is enabled.
Of course, building AppStream without SVG support is not a great idea,
unless the only thing you are using appstream-compose for is non-GUI
applications.
As this is apparently the case for some Flatpak runtimes, we allow
disabling rsvg, but also complain very loudly as soon as we encounter an
SVG icon. This change should simplify early bootstrap and allow building
the complete appstream-compose at a later time in the build process.

Resolves: #434

This was originally implemented in appstream-generator, but apparently
got lost when the code was translated.

This is a temporary change until we can do more testing with GCC 12 to
resolve all sanitizer false-positives.

This helps #395

This helps flatpak/flatpak-builder#418

This helps #395, but the actual proper fix is to implement handling of
screenshot localization in compose properly, as the current code is
completely oblivious to that.
In addition to that change, language compatibility checks for the C
locales also don't seem to work as expected on Alpine, which needs
further investigation as well.

…cies

CC: #395

This fortunately didn't fix any bug, but it's bad style to leave those
uninitialized, as the code around them may always change and thereby
accidentally introduce a crash.

We could maybe also throw a noisy error, but that would be mean to the
CLI tool, and the API consumers will learn to do the right thing anyway
if data turns out to be missing in the result set.

AppStream Generator has this feature, and it made sense to implement it
for ascompose as well. This also simplifies the integration of compose
and asgen.

Possibly the last time I'm doing this, it's really not worth the effort.

Using the version available in Bullseye should be pretty sane by now.
Since more recent releases also have a lot of bugfixes that we want, we
may jump to an even higher version later, to avoid complicated version
checks in the build files.

If an OS has many releases, some EOL information may get lost of we cap
this like all other release information.