fix(deps): drop cilium/ebpf fork, pin real upstream v0.22.0 (#909)

The pinned matthyx/ebpf fork (an intermediate, unmerged snapshot of
cilium/ebpf#1988) carries a global weak-pointer kernel BTF cache: a
race between one gadget's Cache.Kernel() call and a concurrently
loading gadget's Cache.Module() call causes rebaseDecoder to reject a
rebase purely on stale pointer identity. This made trace_network fail
to start intermittently with "apply CO-RE relocations: load BTF for
kmod <mod>: rebase split spec: raw BTF differs", silently leaving
ContainerProfile.spec.egress/.ingress permanently null.

Real upstream cilium/ebpf v0.22.0 already contains #1988's complete,
structurally-correct fix (Cache is fully self-contained per-instance,
no global cache, no weak.Pointer, no identity check to race against),
so this drops the fork replace entirely and pins straight to v0.22.0.

v0.22.0 is several minors ahead and broke two call sites inside the
matthyx/inspektor-gadget fork (btf.NewBuilder's signature, and
VariableSpec's field shape) - fixed there already, so this also bumps
that fork's replace pin to the commit with those compat fixes.

Root-caused and fixed upstream in armosec/private-node-agent#520.


Claude-Session: https://claude.ai/code/session_01PdtsXi1wvbfDu9biC21yRu

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>