Add versioned rule configuration

[sparklkt]

What changed

This PR introduces versioned rule configuration for offline experiments.

• adds RuleConfig and a config loader/hash utility
• adds configs/default.json as the default rule configuration
• moves configurable rule inputs into config-backed fields, including risk keywords, attention thresholds, value field defaults, family mapping, collision lane rules, and pruning thresholds
• threads the loaded config through attention assessment, disturbance scoring, factor generation, collision routing, pruning, and pipeline initialization
• records config_version and config_hash in growth logs and surfaces config metadata in CLI output
• adds tests for default config loading, custom config behavior changes, invalid config failure, and growth-log config persistence

Why it changed

The prototype had accumulated enough hard-coded thresholds and mappings that controlled experimentation was becoming awkward. This PR introduces a versioned config layer so the project can evolve through explicit, reviewable rule changes instead of silent code-only tuning.

Validation

• .venv/bin/pytest -> 24 passed in 0.13s

Safety boundary

• remains offline and rule-based
• no LLM integration
• no network participation in reasoning
• no real external actions
• no bypass of human judgment
• configuration can tune heuristics, but does not grant autonomous external execution or remove high-risk human judgment requirements