Detect handler-induced early termination and populate RunResult.last_active_chain

[proboscis]

Summary

• add RunResult.last_active_chain to the Rust VM result object and public typing surface
• snapshot the VM's latest active chain during exception handling so successful runs can still report the last recovered failure chain
• mark handler protocol early-termination paths as threw before capturing the chain so caught handler failures retain the offending handler provenance

Evidence

1. RunResult always exposes last_active_chain

   • uv run pytest tests/core/test_vm_proto_004_traceback_data.py -q
   • Output: 4 passed in 0.04s
   • uv run python - <<'PY' ... PY
   • Output:
     • ok_last_active_chain []
     • caught_is_ok True
     • caught_traceback_data None
     • caught_last_active_chain_len 5
     • caught_last_effect_result ... {"kind": "threw", "handler_name": "bad_handler", ...}

2. Existing handler failure behavior still passes with the new tracking in place

   • uv run pytest tests/core/test_vm_proto_004_traceback_data.py tests/test_try_finally_in_do.py::test_handler_abandon_raises_error tests/core/test_get_execution_context_effect.py::test_handler_return_protocol_error_also_enriches_execution_context tests/core/test_traceback_spec_compliance.py::test_spec_example_6_handler_return_abandons_inner_chain -q
   • Output: 7 passed in 0.10s

3. Stub surface is type-check clean for the new RunResult attribute

   • uv run pyright packages/doeff-vm/doeff_vm/__init__.pyi
   • Output: 0 errors, 0 warnings, 0 informations

Ref: gh-336-early-termination-detection