Add SRDatalog VulReasoner integration

[StarGazerM]

Summary

• add a self-contained minimal_vulreasoner/ PyReason-to-SRDatalog parity application
• include the original six analyst rules, annotation callback, default GraphML fixture, deterministic stress generator, and exact result checker
• add the executable semantic notebook covering interval encoding, the two aggregation boundaries, non-semiring provenance behavior, and the two-rule lowering
• pin Python 3.10 and all dependencies with uv; use this fork as the editable PyReason oracle and an immutable SRDatalog source archive

Why

VulReasoner's PyReason callback selects one rule-local derivation witness before cross-rule interval merging. The integration preserves that semantic boundary explicitly in SRDatalog and provides a reproducible place to test parity and scaling from the PyReason side.

Validation

• uv sync --project minimal_vulreasoner --frozen
• notebook schema validation with nbformat.validate
• cold PyReason oracle run reaches analystAt(b4)
• bounded differential run on CWE_121_MVP2.graphml: complete temporal AnalystAt maps match exactly
• observed input: 188 nodes, 403 edges, 138 selected connector facts
• SRDatalog: about 48.4 s one-time CUDA build and 16.3 ms run on NVIDIA RTX 6000 Ada

The PyReason cold-start validation is intentionally bounded because first-use Numba cache compilation can take several minutes.