Add evidence-backed zero-day hypothesis engine

[rashedhasan090]

Summary

• Add an offline zero-day hypothesis engine with evidence collection, novelty scoring, rigor scoring, and safe validation metadata
• Integrate the engine into ZeroDayAgent so scans produce defensive hypotheses even when LLM expansion is unavailable
• Normalize optional LLM hypotheses into the same scored schema and add regression tests

Testing

• python3 -m pytest — 25 passed
• python3 -m ruff check wraith_cli/agents/zero_day.py wraith_cli/knowledge/__init__.py wraith_cli/knowledge/zero_day.py tests/test_zero_day.py — passed

Notes

• Full repo lint currently has pre-existing failures outside this change, including unused imports and an undefined ReportGenerator reference in wraith_cli/main.py. Changed-file lint passes.

Slack Thread