Fix scheduler gather scaling and await wait protocol

[proboscis]

Summary

• deduplicate shared interceptor segment tails so current_interceptor_chain() stops rewalking common caller-chain suffixes
• keep scheduler wait bookkeeping O(1) with per-wait counters and add a native WaitEffect path
• align await handling with the CreateExternalPromise -> Wait -> Resume protocol and preserve cooperative sibling cancellation under Gather fail-fast

Verification

• uvx maturin develop --release
• uv run pytest tests/core/test_scheduler_transfer.py -q
  • 9 passed in 1.39s
• uv run pytest tests/core/test_rust_vm_api_strict.py::test_default_handlers_are_module_sentinels_only tests/core/test_sa009_async_handler_spec_gaps.py::TestHandlerImmutabilityContract::test_spawned_child_preserves_full_default_handler_stack tests/effects/test_sync_await_spawn_gather_deadlock.py tests/test_cache_await_spawn_hang.py::TestCacheAwaitSpawnHang::test_no_cache_500_works -q
  • 9 passed in 5.81s
• uv run python - <<'PY' ... N=500 Spawn+Await+Gather benchmark ... PY
  • N=500 elapsed=3.943s ok=True len=500
• cargo check --features python_bridge --tests
  • passed in packages/doeff-vm-core

Acceptance Criteria

• N=500 Spawn+Await(sleep(0.01))+Gather completes well under 10s: 3.943s
• scheduler regression tests pass, including Gather fail-fast sibling cancellation
• single-wait Wait now avoids the Gather(single) path, while WaitMode::Any remains O(1)

Refs: scheduler-gather-quadratic