iouringfs: release MemoryFile allocations on error paths in New

[ibondarenko1]

Problem

iouringfs.New allocates two MemoryFile ranges and then runs several fallible steps. Each error path returns without releasing the allocations:

• mf.Allocate(sqEntriesSize, ...) failing leaks rbfr (the first allocation).
• iouringfd.vfsfd.Init(...) failing leaks rbfr and sqefr.
• the two CacheLineRoundUp overflow checks (EOVERFLOW) leak both.
• iouringfd.mapSharedBuffers() failing leaks both.
• iouringfd.ioRingsBuf.view(...) and .writeback(...) failing leak both.

mf.Allocate returns a refcounted MemoryFile range. The success path transfers ownership to FileDescription.Release:

func (fd *FileDescription) Release(ctx context.Context) {
	fd.mf.DecRef(fd.rbmf.fr)
	fd.mf.DecRef(fd.sqemf.fr)
}

Release runs only once the FileDescription is installed and later released. On any error path in New the FileDescription is never returned, so Release never runs and the allocations leak. New is reachable from io_uring_setup(2).

Change

Wrap the two allocations in a cleanup.Cleanup: register mf.DecRef(rbfr) after the first allocation, mf.DecRef(sqefr) after the second, defer cu.Clean(), and cu.Release() on the success path. The cleanup performs exactly the two DecRef calls that Release does, so it is the correct teardown for every error path, including those after mapSharedBuffers (Release itself does no extra unmapping). This is the cleanup.Cleanup pattern already used across pkg/sentry/fsimpl.

Scope

Hardening. Resource leak confined to the error paths of New; triggering it requires the allocation or a later step to fail.