feat: make async row-group horizon configurable

[eric-tramel]

📋 Summary

Exposes a single advanced RunConfig.max_concurrent_row_groups setting for the async scheduler's fixed row-group horizon. The historical default remains 3; adaptive admission, active-row budgets, and completion/checkpoint changes are intentionally out of scope.

🔗 Related Issue

Refs #741

🔄 Changes

• Adds a validated fixed row-group cap (>= 1) to RunConfig and forwards it to the existing async scheduler setting.
• Documents how the row-group horizon interacts with buffer_size, request concurrency, memory use, and checkpoint timing.
• Adds focused config-validation and builder-wiring tests.
• Keeps adaptive row-group admission, row-budget controls, and completion-state changes out of the public API and this PR.
• Merges the latest main and reconciles the concurrency documentation with the current request-admission terminology.

🧪 Testing

• .venv/bin/pytest packages/data-designer-config/tests/config/test_run_config.py (31 passed)
• .venv/bin/pytest packages/data-designer-engine/tests/engine/dataset_builders/test_async_builder_integration.py (10 passed)
• Ruff check and format check on all changed Python files
• make check-fern-docs (fern check: 0 errors, 1 warning)
• No new E2E coverage required; the scheduler algorithm is unchanged and its existing tests cover fixed-cap enforcement.

✅ Checklist

• Follows commit message conventions
• Commits are signed off (DCO)
• Architecture docs updated

Description updated with AI