Research: lock several lists exactly so functions with 3+ list parameters compile on 3.15t (nbody) #255

User decision (2026-10-17): research option (C) for the nbody regression from #209.

Since #209, a function taking 3 or more list parameters stays interpreted on 3.15t: copy-in to write-back must hold the lists' locks, and CPython's critical-section API covers only 1 or 2 objects (nesting a third suspends the outer one under contention). nbody's offset_momentum (4 lists) and advance/energy (7) went from 31x to 1x.

Options considered: (A) per-list locks only at copy-in and write-back, which can lose another thread's concurrent element writes (a result no CPython interleaving produces); (B) keep refusing (current, identity kept); (C) lock every list exactly, e.g. take each list's internal mutex in address order. The user chose to research (C). Risks to establish: reliance on CPython internal fields (ob_mutex), self-deadlock with re-entrant critical sections, GC stop-the-world interaction.

Until (C) is settled, (B) stays. Done when: a design note states whether (C) is sound on 3.15t (with evidence: TSan / stress tests that fail on a broken variant), and if so it is implemented and nbody compiles again with identity kept.