feat: add browser screenshot baseline support

[Copilot]

Scenarios can now use Vitest browser screenshot assertions without exposing expected baselines to the agent. Screenshots are generated and evaluated in the same Linux/Chromium environment.

Changes

• Baseline isolation

  • Stores baselines under each scenario’s __screenshots__ directory.
  • Copies baselines into the workspace only after agent execution.

• Deterministic screenshots

  • Pins Vitest, Playwright, and Chromium tooling.
  • Uses stable filenames independent of the host platform.

• Regeneration

  • Adds a scenario-local command that regenerates baselines in the evaluation container:

pnpm exec agent-eval --update-screenshots .

• Documentation
  • Documents screenshot assertions, storage, isolation, and regeneration.

[joshblack]

@copilot the way it should probably work for update snapshots is that it should run the scenario exactly as it is within the container since folks likely will change the scenario, take a screenshot, and then only commit the screenshot results as that's the ideal.

[Copilot]

@copilot the way it should probably work for update snapshots is that it should run the scenario exactly as it is within the container since folks likely will change the scenario, take a screenshot, a...

Addressed in a075f55: --update-screenshots now copies the authored scenario into the container as-is, excluding only generated/cache directories and existing baselines, before regenerating screenshots.