refactor(driver-vm): replace OCI registry client with containerd shim

The VM driver's registry image pull and layer-unpack logic was a
hand-rolled OCI client (auth, manifest/index parsing, per-layer blob
download and digest verification) plus a from-scratch tar-layer merge
that only understood plain files, dirs, symlinks, and OCI whiteouts.

Replace it with containerd's own Go client libraries
(core/remotes/docker for registry resolve/fetch/auth,
core/content/local for the content store, pkg/archive for OCI-correct
layer application including opaque dirs, xattrs, and device/fifo
entries). There is no live containerd daemon involved: the new
goshim/ Go module links those packages directly and is built with
`go build -buildmode=c-shared` into a small cgo shared library
(openshell-containerd-shim), loaded at runtime via libloading the
same way this crate already loads libkrun.

- crates/openshell-driver-vm/goshim/: the Go module. Exports
  ContainerdResolveDigest (cheap cache-hit check), ContainerdPullImage
  (resolve + platform-filtered pull into an OCI Image Layout), and
  ContainerdUnpackLayout (apply layers onto a directory).
- src/containerd_shim.rs: Rust-side dynamic loader and safe wrappers,
  mirroring ffi.rs's LibKrun pattern.
- src/driver.rs: removed parse_registry_reference, registry_client,
  linux_platform_resolver, registry_auth, pull_registry_image_rootfs,
  the manual layer-download/whiteout-merge code, and the manual OCI
  layout writer; both the bootstrap-image and prepared-image pull
  paths now call into the shim.
- build.rs / embedded_runtime.rs: embed the shim's compressed shared
  library the same way as libkrun/libkrunfw/gvproxy.
- tasks/scripts/vm/build-containerd-shim.sh: builds the shim (native
  or cross via GOOS/GOARCH/CC), wired into `mise run vm:setup`, the
  Linux CI workflow, and the macOS Docker cross-build (osxcross's
  oa64-clang cross-compiles the shim too).
- Dropped the now-unused oci-client, flate2, and sha2 dependencies.
  Added `go` to mise.toml/mise.lock (new build-time requirement).

Registry auth env vars (OPENSHELL_REGISTRY_USERNAME/_TOKEN) and the
GHCR __token__ convenience default are unchanged. Per-layer pull
progress events are coarser (one PullingImage event instead of one
per layer) since the whole pull now happens inside a single blocking
call into the shim.

Verified against a real registry (docker.io/library/busybox) through
the actual built shared library via Rust libloading, and the macOS
cross-build was verified end-to-end with docker buildx (osxcross
cross-compiling the shim to a real Mach-O arm64 dylib alongside the
Rust binary).

Signed-off-by: Eric Curtin <eric.curtin@docker.com>

Author: ericcurtin

.github/workflows/driver-vm-linux.yml

@@ -165,10 +165,16 @@ jobs:
           OPENSHELL_VM_RUNTIME_COMPRESSED_DIR="${PWD}/target/vm-runtime-compressed" \
             tasks/scripts/vm/build-supervisor-bundle.sh --arch "${{ matrix.guest_arch }}"
 
+      - name: Build containerd shim
+        run: |
+          set -euo pipefail
+          OPENSHELL_VM_RUNTIME_COMPRESSED_DIR="${PWD}/target/vm-runtime-compressed" \
+            mise x -- tasks/scripts/vm/build-containerd-shim.sh
+
       - name: Verify embedded driver inputs
         run: |
           set -euo pipefail
-          for file in libkrun.so.zst libkrunfw.so.5.zst gvproxy.zst umoci.zst openshell-sandbox.zst; do
+          for file in libkrun.so.zst libkrunfw.so.5.zst gvproxy.zst umoci.zst openshell-sandbox.zst libopenshell_containerd_shim.so.zst; do
             test -s "target/vm-runtime-compressed/${file}"
           done