kthread_tests::* leave IF clear before kernel_main_continue(), hanging first x86 registry dispatch

[ryanbreen]

Problem

Every case in kernel::task::kthread_tests::* ends with arch_disable_interrupts(). Nothing re-enables interrupts before control returns to kernel_main_continue(), so that function is entered with IF clear (interrupts masked). This is not theoretical: it permanently hung the first x86 staged-registry dispatch after [STAGE:serial:ADVANCE].

Current workaround (call-site only, not a source fix)

Branch feat/green-tracing @ d9266b71 worked around this at the call site — saving, enabling, and restoring the interrupt flag around the dispatch — rather than fixing the actual test cases. That branch's impl-notes.md records this explicitly as a by-catch discovery: "noted, not rewritten."

What's needed

Fix the source defect: kthread_tests::* cases should not leave the CPU with interrupts disabled on return. Either each case should restore interrupts before returning, or the harness/runner around them should own that guarantee structurally (not via an ad hoc save/enable/restore wrapped around one particular dispatch call site).

Reference

• by-catch discovery citation: impl-notes.md (branch feat/green-tracing @ d9266b71), by-catch entry Update CLAUDE.md: Add critical rule about not pushing to main #6.
• call-site workaround (not the fix) is in feat/green-tracing @ d9266b71.