Server will give a permission denied error.

Without that it takes forever.

The pmbootstrap dir will be used by the new BPO code, so we don't have
a collision.

The new script was quickly hacked together to delete files before
uploading new ones (instead of doing it the other way around as usually)
so we can free up space in case the binary repo mirrors are full.

qt5-qtwebengine is missing for x86, see:
https://gitlab.com/postmarketOS/pmaports/issues/178

Remove the special code for device packages, as it tried to compile
x86_64 device packages for x86 with that. Instead, use the generic
iterate-over-all-packages code above, and add the device packages
there as well.

* kill gpg-agent, too (it gets stuck there when compiling some gpg
  related program and running tests)
* enable armv7 and x86
* only try to build the kernels for a certain arch if there are any
  (armv7 does not have kernels in master right now, prevent a crash
  here)
* only try to build the device packages for a certain arch if there
  are any

This repo will soon be obsoleted by build.postmarketos.org, but here
is the current version I'm using locally anyway.

* increase max retries to 100
* kill cmake if it is still running after failure (that's how the
  kde buildings fail)
* properly call get_packages.py everywhere
* disable luna folder again (re-enabled by accident, since its aports
  aren't fixed yet)

Some KDE packages get stuck with 0% CPU usage when cross compiling with
pmbootstrap (see pmbootstrap#1458). There's a workaround
in pmbootstrap now, which kills processes that do not output anything
for five minutes (postmarketOS/pmbootstrap#1495). This commit changes
the repo build scripts to automatically retry the build in case of
such an error (also in case of other errors, such as download errors).

In addition, there's a bug fix that lets the repo scripts detect x86
packages properly. Previously, we ran grep $arch on the arch= line,
which means it detected x86_64 as x86. Now we have a new
get_packages.py script, that does it properly.