Refactor: WorkerType CHIP→NEXT_LEVEL, consolidate add_worker API, level hygiene

[hw-native-sys-bot]

Summary

• Rename WorkerType enum: CHIP→NEXT_LEVEL, remove DIST (only two kinds of sub-worker: NEXT_LEVEL and SUB)
• Rename scheduler pools: chip_workers→next_level_workers, chip_threads_→next_level_threads_
• Simplify DistWorker::add_worker — clean NEXT_LEVEL/SUB binary split instead of the old CHIP+DIST grouping hack
• Consolidate nanobind API: replace add_chip_worker/add_chip_worker_native/add_chip_process with overloaded add_next_level_worker
• Worker.register() now raises RuntimeError at L2 (was silently useless)
• Extract L2 WorkerPayload unpacking into _run_l2_from_payload() helper

Part of hierarchical runtime refactor (Steps 2-4). See .claude/plans/HIERARCHICAL_RUNTIME_REFACTOR.md.

Testing

• pytest tests/ut/py/test_chip_worker.py — 11 passed
• pytest tests/ut/py/test_dist_worker/ — 17 passed
• ctest (C++ unit tests) — 5 passed
• All pre-commit hooks pass (clang-format, cpplint, ruff, pyright)

[gemini-code-assist]

Code Review

This pull request introduces a validation check in the register method to ensure it is only used at level 3 or higher and refactors the L2 payload execution logic into a dedicated helper method _run_l2_from_payload. I have included a critical review comment regarding the use of run_raw instead of run to prevent potential type errors when handling raw pointers from WorkerPayload.

[gemini-code-assist]

The _run_l2_from_payload helper currently passes raw pointers (integers) from WorkerPayload to self._chip_worker.run(). However, ChipWorker.run() expects higher-level objects, not raw memory addresses. This will lead to a TypeError at runtime. Since WorkerPayload is designed to hold raw pointers for cross-process dispatch, you should use the run_raw method on the underlying _impl instance, which is designed to handle these pointers directly.

        assert self._chip_worker is not None
        self._chip_worker._impl.run_raw(
            payload.callable,
            payload.args,
            payload.block_dim,
            payload.aicpu_thread_num,
            payload.enable_profiling,
        )