Add end-to-end evaluation tests and project documentation

[sezallagwal]

Summary

• Add full pipeline integration tests verifying the complete DSL -> compose -> generate -> validate flow
• Add architecture documentation with Mermaid diagrams covering module map, data flow, and compiler analogy
• Add complete DSL reference with formal grammar, keyword tables, auto-typing rules, and examples
• Add generated project anatomy documentation for consumers of generated servers
• Add security model documentation covering the expression sandbox, AST allowlist, and threat model
• Add testing guide with test suite map, conventions, and instructions for adding tests
• Update README with full project description, capabilities, quick start, and documentation links
• Update CONTRIBUTING with complete setup, development workflow, coding conventions, and PR checklist

Depends on

#14

[dhairyashiil]

Hi Sezal, thanks for adding the docs and evaluation coverage. Could we adjust a few things here so the docs and tests match what the generator really does today?

Could we update the generated-project docs/README claims around tests? The docs currently say generated projects include src/tests/* and an npm test script, but the generator does not emit tests or a test script yet. We should either generate those files/scripts, or remove the claim for now.

Could we also update the DSL reference to include supported constructs like CONTENT_TEXT, CONTENT_IMAGE, and top-level WEBHOOK? The parser supports these, so the reference should include them if this is the main user-facing DSL contract.

For the generated TypeScript validity tests, could we add a stronger smoke test later in this PR or as a near follow-up? Right now the tests mostly check substrings like export / tool; a better signal would be writing a generated project to a temp dir and running tsc or npm run build there.

One stack-level thing: because these PRs are stacked and most target feature branches, the current GitHub CI only shows CLA on most of them. Could we update CI so stacked PRs also run npm run check, not only PRs targeting main? That will help us catch format/lint/test issues earlier while this repo grows.

Overall the documentation direction is helpful, I just want it to stay accurate from the first version.