feat: Optimize schema generation with in-memory parallel processing (53% speedup)

[nathan-stender]

Summary

This PR optimizes the schema generation process that was identified as the major bottleneck in CI test runs.

Problem

• The test_generate_schemas_runs_to_completion test was taking 183-207 seconds (3+ minutes)
• This single test consumed ~35% of the entire test suite runtime
• The bottleneck was sequential processing and heavy I/O operations (temp files, subprocess calls)

Solution

Implemented a new generate_schemas_optimized() function that:

1. Processes schemas in parallel using ProcessPoolExecutor
2. Generates and formats code in memory - no temp files
3. Uses Python APIs directly (black, autoflake) instead of subprocess calls
4. Batches all writes to the end, minimizing I/O operations

Performance Improvements

• Test time reduced from ~180s to 84s (53% speedup)
• CI test suite overall improvement: ~2-3 minutes saved per run
• No functionality changes - just performance optimization

Technical Details

• Added generate_schemas_optimized.py with the parallel implementation
• Original generate_schemas() now defaults to using the optimized version
• Can be disabled with use_optimized=False parameter for debugging
• All existing tests pass without modification

Testing

✅ All tests passing
✅ Schema generation produces identical results
✅ Backwards compatible - original implementation still available

This addresses the most significant performance bottleneck in the test suite.