ggcr: Pulling an image spawns separate goroutines and TCP connections for all layers simultaneously - WithJobs is not respected

[maxb]

Describe the bug

No limits on reasonable numbers of layers to pull simultaneously are applied. Every image layer is handled in parallel, even to the point where failures result because the registry isn't prepared to let one client make large numbers of concurrent requests

To Reproduce

Execute code such as

	desc, err := remote.Get(ref, remote.WithJobs(4))
	if err != nil {
		return err
	}

	image, err := desc.Image()
	if err != nil {
		return err
	}

	cachePath := "/var/tmp/oci-layout"
	ociLayoutDir, err := layout.FromPath(cachePath)
	if errors.Is(err, fs.ErrNotExist) {
		ociLayoutDir, err = layout.Write(cachePath, empty.Index)
	}
	if err != nil {
		return err
	}

	err = ociLayoutDir.ReplaceImage(image, func(desc containerregistry.Descriptor) bool { return true })
	if err != nil {
		return err
	}

whilst monitoring, e.g. with watch -n 0.5 lsof -Pi tcp, how many tcp connections are opened.

Expected behavior

remote.WithJobs(4) should have an effect, bounding how many layers are processed simultaneously.