feat(tutorial): add guided human recording and paced replay

[abrichr]

What changed

• Adds openadapt-flow tutorial --guided for a human-driven demonstration followed by a visible governed replay.
• Adds --presentation-delay SECONDS for a scripted but visibly paced recording and replay.
• Keeps the existing fast tutorial path unchanged for CI and quickstart use.
• Lets the interactive recorder retain an optional read-only system-of-record observation for real effect mining.
• Stops the guided recording only after the final browser event is persisted and the separate system of record shows the saved record.
• Uses the demonstrated parameter value during replay, then runs the unchanged Standard gate and independent verifier.
• Clarifies that the local tutorial has a billable metering class but is not reported or charged.

Why

The existing tutorial completed its scripted recording and governed replay too quickly for a live customer walkthrough. A human-driven recording also shows the demonstration compiler more clearly.

Validation

• Guided-path validation: 7 focused cases passed. The end-to-end case records through the human recorder, retains the final save delta, compiles, certifies, runs Standard, and proves VERIFIED, 2/2 independent effects, and 0 model calls.
• tests/e2e/test_free_path_e2e.py: 12 passed.
• The two final recorder cases pass.
• Ruff passes on all changed Python files.
• Targeted mypy passes for tutorial.py, interactive_recorder.py, and __main__.py.

This is a draft. Do not merge or publish it until the coordinated release owner reviews it.