initramfs: release reserved page range after extract() copies into tmpfs

[sysheap]

At boot, the initrd's physical range is reserved in the page allocator so we don't hand those frames out for anything else:

• Reservation: crates/kernel/src/lib.rs:134 (initrd_range passed to memory::init_page_allocator)
• Consumption: crates/kernel/src/initramfs/mod.rs — find_initrd() turns the range into a &'static [u8] via ValidatedPtr::from_trusted, and extract() walks the cpio and writes every entry into the tmpfs-backed /.

Once extract() returns, the reserved bytes are dead — we've copied everything into tmpfs. For a ~5 MB rootfs.cpio that's ~5 MB of RAM held for the life of the kernel with no consumer.

Why deferred

PageAllocator (crates/mm/src/page_allocator.rs:219) currently exposes only alloc + dealloc. It has no `release_reserved(range)` API. Adding one safely needs:

1. A way to convert a reserved range back into free-list pages, taking the allocator's lock.
2. A lifetime / pointer-provenance argument that nobody still holds the &'static [u8] slice returned from find_initrd() once we release. Today that slice leaks out into extract() and is dropped when that fn returns, but a future refactor that stashes the bytes would silently race. Worth a note / test.

Flagged in the PR review for commit 6760dbd but deliberately not fixed there — this is "not broken, but arguably" territory and the fix is invasive enough to deserve its own changeset.

Rough plan

1. Add `PageAllocator::release_reserved(range: Range<NonNull>)` (or equivalent) that splices the range into the allocator's free structure.
2. At the call site in initramfs::extract(), after successful extraction, call the release API with the reserved range.
3. Unit test: reserve a range, release it, alloc pages and verify the formerly-reserved frames are served.

---

Created by Claude Code