[CAOShurong]

Summary

• recursively convert NumPy scalars and arrays in workflow metadata to BSON-compatible Python built-in values before persistence
• preserve native datetime values and leave the in-memory workflow metadata unchanged
• make the general FireWorks serializer retain NumPy scalar semantics instead of stringifying values such as numpy.bool_
• add nested metadata, BSON encoding, immutability, datetime, and scalar regression coverage

Testing

• python -m pytest fireworks/utilities/tests/test_fw_serializers.py fireworks/core/tests/test_firework.py -q (37 passed)
• repository pre-commit hooks on all four changed files
• non-MongoDB portable subset (94 passed, 56 skipped, 4 deselected); the excluded local-Windows files require Unix cat/echo/touch commands or a running MongoDB, and representative failures reproduce on the exact base commit
• sdist and wheel build
• clean-venv install of the wheel followed by nested numpy.bool_/integer/float/array Workflow.to_db_dict() and BSON.encode() round-trip probe
• git diff --check

Closes #522

Disclosure: This contribution was developed with OpenAI Codex assistance. I reviewed the implementation and regression coverage.