Add critical user flow test gate

[iatopilskii]

Summary

Adds a dedicated Playwright release gate for DevHub’s most critical developer and coding-agent user flows.

The suite runs against a local production build. It covers desktop Chromium and WebKit plus explicitly named Playwright-emulated Pixel and iPhone profiles. No production monitoring, scheduled checks, or alerts are added.

Why

DevHub already had broad Vitest and Chromium regression coverage, but there was no isolated cross-browser gate for journeys whose failure prevents a developer or coding agent from getting useful content.

Critical flows covered

• Home page → fetch and copy a complete bootstrap prompt.
• Desktop Cmd/Ctrl+K or mobile-menu search → open Databricks CLI documentation → copy useful Markdown.
• Templates search → open SaaS Subscription Tracker → copy the agent prompt → validate the Replit handoff.
• Simulated clipboard failure → display the retry state → recover without reloading.
• Validate coding-agent entry points:
  • /api/bootstrap-prompt
  • /llms.txt
  • Raw documentation
  • MCP tools/list
  • MCP calls to list_docs_resources and get_doc_resource

Implementation

• Adds playwright.critical.config.ts and a documented P0/P1 test plan.
• Adds pnpm test:critical and an independent critical-flow CI step.
• Uses the real browser Clipboard API with explicit permissions in desktop Chromium; deterministic clipboard behavior remains for cross-engine tests and controlled failure injection.
• Verifies required MCP tools as an extensible subset and calls the tools to validate useful responses.
• Names device-profile projects emulated-pixel-chromium and emulated-iphone-webkit so CI output cannot be mistaken for physical-device coverage.
• Refuses to reuse an unrelated local server for the critical production-build gate.
• Preserves screenshots, traces, and videos for failed critical tests as short-lived CI artifacts.

Validation

• Critical suite: 17 passed, with 3 browser-independent API duplicates intentionally skipped.
• MCP tests: 13 passed.
• pnpm build: passed.
• pnpm typecheck: passed.
• Prettier check: passed.
• Content validation: passed.
• Image verification: passed.
• fallow dead-code: no issues.
• Manual browser and SEO smoke checks: passed with no runtime errors; title, H1, and canonical remain correct.

Existing baseline failures

The broader regression suite still reports unrelated failures in files not changed by this PR:

• Vitest: 311 passed; 2 AppKit documentation broken-link tests failed.
• Existing Chromium Playwright suite: 182 passed; 1 layout-height assertion failed by 6 px after the latest AppKit docs sync.
• fallow dupes reports existing duplication groups in untouched production files.

Test plan

See tests/critical/README.md for priorities, expected outcomes, browser coverage, and local commands.