Collect some basic system information about the runners the CI jobs
run on. This is done iwth the intention to have diagnostics information
in advance in case builds fail spuriously in some instances.

Signed-off-by: David Bauer <mail@david-bauer.net>

Despite what GitHub states in their documentation, the runners feature a
different core count. Automatically set the build thread-count by the
amount of available CPU cores.

Link: https://docs.github.com/en/actions/using-github-hosted-runners/about-github-hosted-runners/about-github-hosted-runners#supported-runners-and-hardware-resources



Signed-off-by: David Bauer <mail@david-bauer.net>

It makes sense to run CI build-tests in case the CI itself is altered.

Signed-off-by: David Bauer <mail@david-bauer.net>

* add missing libelf-dev dependency (to build x86-64)
* remove subversion dependency
* add rsync dependency
* add ecdsautils to docs

to prevent word splitting and globbing

The scripts directory contains most of our build system, so changing it
should trigger a build of all targets.

This has become possible with OpenWrt 21.02 and while the Dockerfile
already received that updated, we forgot to update the CI and the
documentation.

Currently we do not perform CI firmware builds on the next-2102 branch.

Build Gluon for all branches starting with "next" to increase the
coverage of our build tests.

Signed-off-by: David Bauer <mail@david-bauer.net>

apt does not have a stable CLI interface. Don't use it in scripts.

This reverts commit d9621048.

ubuntu-latest is now assigned to Ubuntu 20.04. As we use custom apt
sources for 18.04, pin to this version for now to fix the CI.

The performance benefit the cache brought was due to a broken sources
CDN mirror handling in OpenWrt.

The cache brings no measurable performance benefit. Disable it to slim
down the pipeline steps.

This adds the ability to cache OpenWrt dependencies on a per-target
base. Artifacts over 10MB are excluded, as GitHub imposes a limit of 5G
of available space per repository cache. This affects mostly
linux(-firmware) and gcc / gdb.

The goal is to reduce the total amount of requests necessary to fetch
dependencies.

- Rely on shebang instead of setting shell in workflow
- Run whole install-dependencies.sh in sudo
- Use /bin/sh instead of bash
- set -e

The new step is added with `if: ${{ !cancelled() }}`, so the logs are
stored even when the build fails.

This enables build-testing only on master as well as next and release
branches.