render: dispatch InputSubmitted and ChoiceSelected with Source; re-ground ChoiceSelected to Values []string

[joestump]

What

Resolves the two standing TODO(a2tea) markers in event/event.go (part of epic #18): InputSubmitted and ChoiceSelected are now actually dispatched by the renderer, both carrying Source{ComponentID, SurfaceID} like ButtonClicked.

How

• InputSubmitted — Enter on a focused TextField (or DateTimeInput, which shares the TextField edit path) emits event.InputSubmitted{Source, ID, Value} via a new Surface.submitInput cmd. Value follows the same shadowing rule as rendering/FieldValues: the edited text when the user has typed, else the literal / resolved-binding seed (editSeed), so an unresolved {binding} placeholder submits "" and never leaks. Enter on buttons, modals, and checkboxes is untouched.
• ChoiceSelected — re-grounded to Values []string per the package doc's re-grounding note before wiring (ChoicePicker is natively multi-value; changing an emitted type later is worse). togglePickerOption now returns a tea.Cmd: it compares the normalized (option-declaration-order) selection before and after the toggle and emits the full post-toggle selection only when the set changed — a radio-semantics Space on an already-selected single-select option emits nothing.
• Package docs updated: event drops the "not emitted yet / provisional shape" caveats; render's package and Update docs describe the new dispatches. The README package blurb and the README + docs/wire-format.md "not yet" lists drop the events line.

Tests

• render/textfield_test.go: Enter on an unedited field submits the literal with Source set and no ButtonClicked; Enter after typing submits the edited value; Enter on an unresolved bound field submits "" (no placeholder leak).
• render/inputs_test.go: Enter on DateTimeInput emits InputSubmitted; single-select Space emits ChoiceSelected{Values: [b]} and re-selecting emits nothing; multi-select toggles emit the full selection in option order, down to the empty selection.
• event/event_test.go: shape pin updated for Values []string.
• Gate: go build ./... && go vet ./... && go test ./... all green.

Notes

• website/ docs are deliberately untouched — deferred to the post-merge docs pass, per the working agreement. The stale "editing beyond TextField is read-only" lines in README/docs/wire-format.md (obsoleted by Editable CheckBox / ChoicePicker / Slider / DateTimeInput (finish #15) #42) are likewise left for that pass; only the events line this ticket owns was dropped.

Fixes #43

🤖 Posted on behalf of @joestump by Claude.

[joestump]

Adversarial review of the diff against issue #43. Gate re-run locally: go build ./... && go vet ./... && go test ./... -count=1 all green.

What I verified

• State/guard safety. submitInput (render/render.go:419) and togglePickerOption (render/inputs.go:116) can only run when the focused component actually exists and is the right kind — focusedIsTextEditable/focusedIsChoicePicker go through focusedComponent, which returns a zero Component for an ID an Apply merge removed, so the missing-component path is unreachable and the remaining nil-map reads (s.fieldValues[id] on a nil map) are safe Go. pickerCursor clamping still covers options shrinking across merges.
• Key routing. Enter ordering (modal → button → text-editable → checkbox) inserts the new branch without stealing anything: buttons still activate, checkboxes still toggle, Space on a text field still types a space, and the ChoicePicker Space branch now correctly propagates the cmd. Traced every other key case — no regressions.
• ChoiceSelected multi-value correctness. Values is []string in option-declaration order; the before/after compare in togglePickerOption normalizes the pre-toggle selection with the same option-order walk, so radio-semantics re-select genuinely emits nothing while a real change (including clearing to the empty selection, which emits a non-nil empty slice) always fires. Verified the single-select, multi-select, and undeclared-literal edge paths by hand.
• InputSubmitted value shadowing. Edited text wins; unedited falls back to editSeed, so an unresolved {binding} placeholder submits "" and never leaks — matches the rendering path.
• Tests assert behavior, not execution. The new tests pin Source (ComponentID + SurfaceID), payload values, option ordering, the no-event radio case (cmd == nil), the empty-selection event, the edited-vs-literal distinction, and the no-ButtonClicked invariant. event_test.go shape pin updated for Values []string.
• Ticket compliance. Both TODO(a2tea) markers gone, package doc drops the provisional-status warnings (grep confirms no leftover "not emitted"/"provisional" markers outside historical wire-format prose), README + docs/wire-format.md drop the events line, website/ untouched per the working agreement, both events carry Source like ButtonClicked.

Blockers

None.

Nits

1. render/render.go:414-418 — submitInput doc overclaims equivalence with FieldValues. The comment says Value follows "the same shadowing rule as rendering and FieldValues", but gatherFieldValues skips bindings entirely (resolveDynamicString returns nil for non-literals) while submitInput resolves them via editSeed against s.data. On a surface with an applied data model, Enter on an unedited bound field emits InputSubmitted.Value = <resolved value> while FieldValues() omits that key. The event behavior is arguably the more correct of the two (it matches rendering), so no code change requested — but the comment should say "same rule as rendering" and drop the FieldValues claim, or FieldValues should learn the same resolution in a follow-up.
2. README.md:104-105 — internal contradiction left in a touched file. The "not yet" list still says CheckBox/ChoicePicker/Slider/DateTimeInput "remain read-only visuals" three lines below a blurb (edited by this PR) describing ChoicePicker emitting ChoiceSelected on selection change. The PR body defers this to the docs pass, which is defensible, but it's a one-line fix in a file this diff already edits.
3. render/inputs.go:149 — silent normalization without an event. When a picker's literal names undeclared options (e.g. literal ["x","a"], Space on already-selected a), choiceValues is rewritten to the normalized ["a"] — so FieldValues() changes — while before == out suppresses the event. The comment documents this as deliberate ("dropped by normalization, not by the user's toggle") and it only bites on producer-error input, so noting for the record only.

Verdict: approve. The nits are documentation-precision issues; the dispatch logic, payload shapes, and tests are correct.

🤖 Posted on behalf of @joestump by Claude.