refactor: widget submit script switches on data-collect, not ftype (0.5.0 cleanup batch)

[silversurfer562]

Summary

Fourth 0.5.0 cleanup item, chair-ruled into its own PR (2026-08-16): the widget submit script's reader is no longer a per-ftype if/else — each rendered field carries a data-collect attribute (value | checked-one | checked-many | rulings | ranked | rulings-with-text), emitted at render time from the per-type _COLLECT_MODES map, and the script switches on that attribute. A new construct type that answers like an existing one (one checked radio, per-row rulings, a ranked list…) registers its mode at render time and needs no script edit.

• The rulings readers and the required-row gate scope rows by [data-item] instead of construct CSS classes (.ae-triage-row / [data-assume-row])
• Number coercion keys off el.type === 'number' (the control, not the construct)
• list_style SINGLE_SELECT emits checked-one at render time — replaces the reader's runtime radio sniffing
• Posted payloads are byte-identical — this is a reader-plumbing refactor only

Lockstep updates

• Round-trip simulator (tests/test_widget_roundtrip.py): _submit mirrors the same data-collect switch; new list_style single-select round-trip regression
• Drift catcher repinned: test_script_switch_covers_the_emitted_collect_modes asserts the emitted modes AND the script's literal cases both equal the pinned vocabulary (was: ftype special-case coverage)
• Nine construct tests repinned from "ftype === '<type>'" script text to data-collect emission / mode === '…' cases (confirm's "nothing pre-selected" pin tightened to the space-prefixed checked attribute so checked-one isn't a false hit)
• widget-preview lane honors PORT (autoPort: true) so parallel sessions can each run their own preview

Verification

• 622 passed (was 611 + the new regressions); ruff + black clean on touched files
• Browser receipt: reference form (all 8 constructs + 7 base types) filled and submitted in the widget-preview lane with real clicks (including ranking remove/add and the assumption-review edit reveal); the captured sendPrompt payload carries the sentinel + all 15 answers and validates through collect_form_response end-to-end

After merge

Re-sync the attune-ai mirror copy of test_widget_roundtrip.py (import-block swap only, per the attune-ai #2073 pattern).

🤖 Generated with Claude Code