sysext/containerd: read drop-ins from /etc/containerd/conf.d

[satwiksps]

Changing containerd settings currently means copying the default config. That copy can miss updates to the defaults later.

This change makes both shipped configs read /etc/containerd/conf.d/*.toml, so users can keep their own settings in separate files.

Fixes flatcar/Flatcar#1127.

How to use

Create /etc/containerd/conf.d/50-log.toml with these contents, then restart containerd:

version = 2
[plugins."io.containerd.grpc.v1.cri"]
max_container_log_line_size = 65536

Use version = 2 to match the shipped configs. Files are read in filename order. Lists are combined, and some empty or zero values do not replace existing settings. A missing or empty directory leaves the settings unchanged. If you use your own main config, add the imports line there too.

Testing done

python3 .repair-validation/containerd-probes.py "$PWD" \
  "$RUNNER_TEMP/containerd-bin/bin/containerd" "$RUNNER_TEMP/containerd-probes"
bash .repair-validation/flatcar-vm.sh "$PWD" "$RUNNER_TEMP/flatcar-vm"
git diff --check

• All 22 Linux config checks passed with containerd 2.3.4.
• Both configs passed service tests on Flatcar 4790.0.0 with containerd 2.2.5. Tests covered missing and empty directories, overrides, invalid config and recovery.
• Docker and CRI containers ran successfully. The CRI container exited with code 0, and the running service used the requested log limit.
• git diff --check passed.

Test logs and scripts. The tests used the same config files as this PR. The VM used a published Flatcar image, cgroup v2, SELinux permissive mode and host networking. SELinux enforcing mode, CNI, Kubernetes and a custom OS build were not tested. The test scripts are on a separate branch.

• Changelog entry added.
• Inspected CI output for image differences.

AI helped with the investigation, code change and tests.