fix(fc): reflink-copy base rootfs image instead of full byte copy

DebugfsRootfsPreparer.Prepare did a plain io.Copy of the entire base
rootfs image on every Deploy, dominating `onctl up` time for the fc
provider (multi-second sys time copying a large image). The cache
disk preparer already avoided this cost via a CoW reflink; the rootfs
path now does the same, falling back to a full copy on filesystems
without reflink support.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PVpF4aLeq6xpokgHiBCy4v

Author: cdalar

internal/providerfc/common.go

@@ -464,7 +464,7 @@ func (DebugfsRootfsPreparer) Prepare(baseImage, destPath, sshPublicKey, username
 	if baseImage == "" {
 		return errors.New("fc.rootfsImage is not configured")
 	}
-	if err := copyFile(baseImage, destPath); err != nil {
+	if err := copyRootfs(baseImage, destPath); err != nil {
 		return fmt.Errorf("failed to copy base rootfs %q: %w", baseImage, err)
 	}
 	if sshPublicKey == "" {
@@ -595,6 +595,20 @@ func reflinkCopy(src, dst string) error {
 	return nil
 }
 
+// copyRootfs creates dst as a copy of src, preferring a copy-on-write
+// reflink (near-instant and disk-space-free regardless of image size) and
+// falling back to a full byte-for-byte copy when the host filesystem
+// doesn't support reflinks (e.g. plain ext4, or non-Linux). Every Deploy
+// calls this once per microVM, so on a CoW filesystem (btrfs, or XFS with
+// reflink=1) it turns what was previously a multi-second copy of the whole
+// base rootfs image into a near-zero-cost metadata operation.
+func copyRootfs(src, dst string) error {
+	if err := reflinkCopy(src, dst); err == nil {
+		return nil
+	}
+	return copyFile(src, dst)
+}
+
 func copyFile(src, dst string) error {
 	in, err := os.Open(src)
 	if err != nil {