[P2] No test sandbox or record/replay for CI-friendly testing

Problem

All development and testing currently runs against production Glean. There is no sandbox instance, no seed data, no record/replay, no mock server, no dry_run flag.

Current state

The toolkit uses VCR cassettes internally for its own test suite, but there is no user-facing testing story.

Evidence

From PM's 3P Agents Gap Spec:

• 2 days, ~50+ API calls all hit production with real enterprise data
• CI/CD pipelines can't test the knowledge layer
• Integration tests either hit production (flaky, slow, requires secrets) or mock at HTTP level (fragile, breaks on API changes)

Fix

Consider:

• Record/replay proxy that captures real responses for offline testing
• A dry_run flag returning structurally-valid fake responses
• Mock service or test fixtures that users can use in their own CI
• Document the VCR cassette approach as a pattern for users

Priority

P2 — Blocks robust CI/CD for agent builders.

Source: 3P Agents Stack Rank gap spec, item 4.2

Start by reading the toolkit's existing VCR cassette-based test suite and compare its approach with the proposed record/replay proxy, dry_run flag, mock service, and user-facing fixtures. Decide which testing story is in scope before identifying the relevant implementation and documentation entry points. Done means users can run reliable offline tests in CI without production secrets or enterprise data.