test(uffd): unregister UFFD range before fd close in cross-process cleanup

[ValentaTomas]

Summary

Split out from the feat/free-page-reporting branch as a small, standalone preparatory PR.

• Adds unregister() helper in fd_helpers_test.go that wraps UFFDIO_UNREGISTER.
• Calls it from the late `cmd.Wait()` cleanup in `configureCrossProcessTest`, before the early `uffdFd.close()` cleanup runs (cleanups are LIFO).

Why now (and why it's safe to land alone)

Today this is a no-op: no test enables `UFFD_FEATURE_EVENT_REMOVE`, so the kernel never queues REMOVE events that could keep `munmap` blocked on un-acked events against a still-registered range.

Once the follow-up REMOVE-handling PR enables that feature, missing this cleanup would manifest as flaky hangs in test teardown rather than as an obvious failure. Landing the cleanup first means that follow-up PR doesn't silently bring along a behavioural test-cleanup change alongside the production logic.

Diff

`+23 / -0` across 2 files, both `_test.go`.

Test plan

• `go build ./...`
• `golangci-lint run ./pkg/sandbox/uffd/userfaultfd/...` — 0 issues
• `go test ./pkg/sandbox/uffd/userfaultfd/... -count=1` — pass

[cursor]

PR Summary

Low Risk
Low risk: test-only cleanup change, but it introduces an extra UFFDIO_UNREGISTER ioctl that could fail on unsupported kernels or if the range/FD state is unexpected during teardown.

Overview
Ensures cross-process userfaultfd tests explicitly call UFFDIO_UNREGISTER to tear down the registered address range during cleanup (before the FD is closed), preventing potential teardown hangs from queued/unacked REMOVE events in future tests that enable REMOVE notifications.

^{Reviewed by Cursor Bugbot for commit a0522c3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[claude]

LGTM — test-only, no production code changes, and the cleanup ordering logic is correct.

Extended reasoning...

Overview

Changes are confined to two _test.go files in packages/orchestrator/pkg/sandbox/uffd/userfaultfd/. Adds an unregister() helper wrapping UFFDIO_UNREGISTER (mirroring the existing register() pattern) and calls it in the LIFO cleanup chain before uffdFd.close().

Security risks

None. These are test helpers with no impact on production paths, authentication, or data handling.

Level of scrutiny

Low. The diff is +23/-0 in test files only, mechanically mirrors the existing register() implementation, and is currently a no-op (no test enables UFFD_FEATURE_EVENT_REMOVE). The LIFO ordering reasoning is sound and well-documented.

Other factors

No bugs found, no outstanding reviewer concerns to address. The preparatory-PR rationale (isolating cleanup logic from the follow-up REMOVE-handling change) is sensible engineering hygiene.