fix(vmm): use the real FICLONE ioctl number

[WaylandYang]

chain::reflink_copy passed 0x40209409 as FICLONE. The real number is _IOW(0x94, 9, int) = 0x40049409. The kernel matches the full ioctl number, including the size field, so it answered ENOTTY. The fallback treats ENOTTY as "this filesystem has no reflink", so every call silently fell back to a full streamed copy: chain-assembly base memory, and the bake's rootfs baseline clone. That happened even on btrfs, XFS, and ZFS 2.2+.

Found by @jrimmer while measuring #321 (described in a comment there, but the change wasn't in that branch), so it lands on its own here.

Verification

On Ubuntu 22.04 / kernel 5.15, using an XFS loopback made with reflink=1 and a 200 MiB random source file:

0x40209409 -> ENOTTY
0x40049409 -> clone OK
df used: 240 MiB for src + clone (extents shared); cmp: content identical

A unit test pins the constant to its _IOW derivation.

🤖 Generated with Claude Code