test: split e2e tests into integration + e2e pairs

[planetf1]

Context

Part of the Testing Infrastructure & Strategy Overhaul epic (item 2a). Depends on granularity markers (#727) and marker audit (#728).

Objective

Audit current e2e tests and create mocked integration variants for CI and local dev, keeping the original e2e tests for nightlies. This follows the agreed "split, don't just downgrade" principle — both variants may diverge over time to focus on their respective concerns.

The pattern to follow is #567 (test_astream_mock.py).

Work

• Inventory all current e2e tests and identify candidates for integration variants
• Create mocked integration variants for each candidate (using the test_astream_mock.py pattern)
• Ensure integration variants run in PR CI and local dev (<15 min budget)
• Ensure original e2e tests continue to run in nightlies
• Apply correct granularity markers (integration / e2e) to both variants per test: define and implement granularity marker taxonomy #727
• Update contributor documentation on when to create both variants vs. a single test

Notes

• This covers both test/ and example scripts
• Do this before 2b (parametrisation) — it delivers immediate CI coverage value and builds understanding of the test landscape
• Integration variants should use mocked backends, not just lighter models

[planetf1]

Additional work has been done in 813 'bottom-up' to improve coverage.

A task for a future iteration is to see if any gaps in integration/unit testing can be filled by a variation of what we have in e2e currently.

[planetf1]

Partial progress note (Aug 2026):

Bottom-up unit/integration coverage work landed in #813 and the marker taxonomy/audit work closed in #742. This improved overall coverage significantly.

However, the explicit e2e → integration split described in the checklist above has not been done systematically — the inventory of e2e candidates, mocked-variant creation, and contributor documentation updates remain. Carrying forward as backlog.