The format of cgroup path is different for systemd and cgroupfs
drivers.

For systemd driver, it's

```
0::/system.slice/docker-d4ea646fc22c701dbb146e52db4e9125dcca2eebb2f5552f90fedbb28a0f0716.scope
```

For cgroupfs driver, it's

```
0::/docker/2cf2e0be458a80acb354c953f7bb03de5d2d277dfcb8ebaa6575f95668a0c15f
```

Handle both formats.

And only if necessary.

… is big enough

Before this change, we were trying to handle offsets by doing pointer
arithmetics (like `str + offset` where `str` is `char*` and `offset`
an integer).

Verifier in kernels <6.0 was not happy with that and was complaining
about unchecked bounds.

Fix that by making `buffer_append_str` aware of offsets and doing all
the bound checks there.

* It's cleaner.
* Using `char` array was triggering verifier errors on kernel 6.0.