Add NLDataQuerySkill - natural language to SQL query bridge (#1 priority from MEMORY)

[lbartoszcze]

Summary

• NLDataQuerySkill - Translates plain-English data questions into SQL queries and executes them as a premium paid service ($0.015/query, 50% markup over raw SQL)
• Bridges NaturalLanguageRouter's intent understanding with DatabaseRevenueBridge's paid query execution
• 6 actions: query (NL→SQL→execute), explain (show generated SQL without running), discover (schema introspection), suggest (auto-generate example questions from schema), teach (learn custom NL→SQL mappings), stats (revenue tracking)
• SQL generation uses keyword matching, aggregate detection (SUM/AVG/COUNT/MAX/MIN), GROUP BY inference, ORDER BY/LIMIT detection, time filters, WHERE clause extraction
• Schema caching with table/column type awareness for intelligent column selection
• Learned mappings persist across sessions for self-improvement
• Revenue tracking: per-customer, per-query, with full audit trail
• Read-only enforcement: only SELECT/WITH/EXPLAIN/PRAGMA allowed
• Registered in autonomous_agent.py DEFAULT_SKILL_CLASSES
• 18 new tests, all passing. 17 smoke tests still pass.

Pillar: Revenue Generation (primary), Self-Improvement (supporting)

This is the critical missing piece in the revenue pipeline. Without this, customers must know SQL to use paid data services. With this, any user can ask "show total sales by region" and get results — dramatically lowering the barrier to paying for data analysis services. The teach action enables self-improvement: the system learns better NL→SQL mappings from operator feedback.

Test plan

• 18 unit tests covering tokenization, schema discovery, SQL generation (aggregates, counts, selects, limits), all 6 actions, error handling
• 17 smoke tests pass (no regressions)

🤖 Generated with Claude Code