Improve validator performance at scale

[satyakigh]

Summary

This change significantly improves performance for applications that synthesize large numbers of templates.

• Enabled concurrent Rego validation by replacing shared mutable evaluation state and the global lock with per-thread evaluation contexts.
• Added bounded per-thread caching for compiled Rego regex patterns.
• Skipped globally filtered built-in rules before evaluation using guarded Rego clauses.
• Optimized intrinsic-source lookups, reference-graph iteration, and intrinsic-name suggestions.
• Prevented cross-template leakage of custom rule descriptions, schema scenario filters, regions, models, and suppression state.
• Hardened shared pattern caches against poisoned locks.

Overall performance

Engine	Before	After	Improvement
Rego	178.86 s	78.80 s	2.27x faster
CEL	39.74 s	32.89 s	17.2% faster

Concurrent validation through one shared Rego engine improved from 9.29 seconds to 3.26 seconds with four workers, a 2.85x speedup.