feat: add reusable dataset generation library API while preserving CLI

[owenqwenstarsky]

Motivation

• Provide a programmatic API so the tool can be imported as a library and used to generate datasets without the CLI while keeping the existing CLI behavior intact.

Description

• Added exported types GenerateDatasetOptions and GenerateDatasetResult and a new generateDataset(options) function that encapsulates prompt reading, concurrency, OpenRouter calls, JSONL writing, spend tracking, and key cleanup.
• Refactored the existing generation flow out of main() into generateDataset() and updated main() to parse CLI args/env then call the new function so CLI behavior is preserved.
• Made stderr injectable via the options to allow library consumers to control output and progress rendering, and added a small refactor to use ensureReadableFile() inside the new function.
• Added an end-to-end test that mocks fetch and validates JSONL output and counts, and bumped lockfile root version to match package.json (0.1.12).

Testing

• Attempted npm test with the default test runner invocation (failed in this environment due to how the test glob is invoked), and then ran a targeted run: npm run build:test && node --test dist-test/test/datagen.test.js which executed the test suite and all tests passed (18 tests, 0 failures).
• Ran npm run build which succeeded and produced compiled outputs.
• Added and ran a new test generateDataset writes JSONL output for prompts that validates output structure and counts and it passed in the test run.

---

Codex Task