Canonicalize + type AI-automation summary rows (response side)

[mocha06]

Part of the "Typed payloads: parse, don't validate" milestone. Follows #416 / PR #417.

Problem — response side, not input side

_is_ai_automation_summary_row (ai_preflight.py) and cli/commands/ai_automation.py read listing rows with dual-alias branches:

• ai_preflight.py: row.get("action_id") or row.get("actionId"), row.get("action_params") or row.get("actionParams")
• cli/ai_automation.py: row.get("action_params") or row.get("actionParams"), action_params.get("aiParams") or action_params.get("ai_params"), ai_params.get("fieldIds") or ai_params.get("field_ids"), row.get("event_id") or row.get("eventId")

Unlike the input-side issues, the row shape is decided by our own query documents, not the API input schema. A dual-read means two query paths produce different key styles (e.g. one query aliases eventId: event_id, another does not).

Approach

1. Find the query document(s) that produce these rows and canonicalize the selected key style (pick one — align with how the API returns them / how the SDK's other agent queries alias).
2. Type the row with a small read-model (or reuse automation_graphql_types.py), then read typed attributes.
3. Delete the dual-reads.

Do the query alignment first; typing the row on top of two disagreeing queries just moves the problem.

Acceptance

• One canonical key style for AI-automation summary rows; dual-reads removed.
• uv run pytest -m "not integration" + ruff + import-linter green.