[sergiomanzur]

An unresolved tail from an AOT subroutine running inside NMI/IRQ currently enters a fresh interrupt interpreter before checking whether an existing interpreter owns the continuation. That nested run consumes the caller's continuation and RTI, then returns to compiled code with a guest stack that has already unwound.

Check the interpreter bounce owner first, and only select interrupt-root execution when there is no paired RTS/RTL return frame (hrv == 0). Near and far subroutines retain the existing balanced fallback, including indirect-dispatch handling.

Validation:

• Added a ROM-free fixture linked against the production interpreter and bridge. On unchanged main it reports 10 failed checks: incorrect S, double-executed continuations, and missing owner unwind for both JSR and JSL.
• python -m pytest tests/test_interrupt_tail.py -q: 2 passed, covering the new fixture and the existing bridge contract harness.
• New fixture also verifies interrupt-root RTI and enclosing interrupt-scope preservation.
• Independent code review found no actionable issues.

This was extracted from ISSD Native bring-up and adapted to current main (0788387). No commercial ROM, generated game source, frame scheduling changes, or game-specific addresses are required. Cross-title gameplay qualification is left to the maintainers; this PR establishes the synthetic ownership contract.

[mstan]

Reviewed ownership ordering: interpreter bounce ownership must win, and only an unpaired interrupt root may run to RTI. Near/far subroutines retain balanced fallback. The production bridge regression and existing bridge harness pass.

Validation: five-head integration passed 826 Python tests (2 skipped) and 109 Rust tests. After excluding #198 for its separate title-save compatibility blocker, the four-head combination passed all 69 focused tests. Linux, Windows, macOS and Lua CI are green for this reviewed head. No new cross-title gameplay playtest is claimed.