Add word cloud poll type

[junkerderprovinz]

Summary

A minimal take on #79's word cloud request.

What this adds

A "Word Cloud" option next to "Choice" when creating a poll. Attendees type their own word or short phrase instead of picking from choices the moderator set up ahead of time. There's nothing to configure beyond the title, so the options list, "Add" button and "Multiple answers" are hidden for this type in the creation form.

Design choice: reuses poll_opts/poll_votes, no new schema

Polls.submit_word/4 looks for a poll_opt on the poll whose content matches the submitted text case-insensitively (trimmed), increments its vote_count if found or creates a new poll_opt at vote_count 1 if not, then records a PollVote exactly like Polls.vote/4 does. That means "have you already answered this poll" (Polls.get_poll_vote/2), percentages (Polls.set_percentages/1) and the :poll_updated broadcast all work unchanged for both poll types.

The attendee's own screen renders the cloud through PollComponent. The shared/projected presenter display (/e/:code/presenter) has its own separate markup that doesn't use PollComponent, so it needed its own word-cloud branch; both now compute font size through a shared ClaperWeb.Helpers.word_size/3 (extracted from what was a PollComponent-private function, now takes an optional base/range so the bigger presenter screen can size its words differently than the moderator's small panel).

The cloud itself is plain CSS (font-size scaled by each word's share of total votes). Didn't bring in a word-cloud JS library for what's fundamentally a sized-text layout; happy to swap in something fancier if you'd rather have real cloud packing and rotation, and can point me at a preferred library.

New migration adds type (string, default "choice") to polls; every existing poll implicitly stays :choice, no behavior change for polls that already exist.

Defects found and fixed on review

A second pass over my own branch turned up two real bugs, both now fixed with regression tests.

submit-word accepted a submission against whatever interaction was on screen. The handler passed socket.assigns.current_interaction.id straight into Polls.submit_word/4 without checking what that interaction actually was, and the event is client-triggered, so an attendee could fire it regardless of what was being shown. With a choice poll active, upsert_word_poll_opt/2 inserted the submitted text as a brand new PollOpt on that live poll, which is arbitrary text injection into a running presentation. With a form, quiz or embed active, the id belonged to a different table entirely, and because PollOpt.changeset/2 declares no foreign_key_constraint, the insert hit poll_opts_poll_id_fkey, raised Ecto.ConstraintError and took the LiveView process down. The handler heads now match %Polls.Poll{type: :word_cloud, enabled: true} and the event is ignored otherwise, the same shape this module already uses to guard save on chat_enabled and react/unreact on message_reaction_enabled.

The word cloud ignored the "show results" toggle. The cloud rendered as soon as an attendee had submitted, whatever the moderator's setting, so a submitting attendee saw the live cloud even while results were meant to stay hidden. It also rendered directly beneath the "thanks, submitted" message, showing both at once. It is now gated on @show_results alone, which is what the choice branch of the same component already does (bar width zeroed and percentages hidden unless @show_results).

Translations

Ran mix gettext.extract --merge, so default.pot and every locale's default.po pick up the seven new strings. They land untranslated and fall back to the English source until a translation pass, matching how strings arrived in previous feature PRs such as #214. Note that the same run also refreshes #: source references across the catalogues, which had drifted from the current source tree before this branch, so the diff on those files is larger than the seven new strings alone.

Not included

Kept this to the poll mechanics themselves. A few things deliberately left out:

• profanity or spam filtering on submitted words
• a moderator-side control to remove a single word from the cloud (deleting the whole poll still works)
• converting an existing Choice poll (with prior answers) to Word Cloud via the edit form doesn't clear its old poll_opts, so stale choice options would resurface as pre-existing "words"; worth a follow-up if that path matters to you
• no locking against two attendees submitting the same brand-new word at nearly the same instant, which could in theory create two near-duplicate poll_opts instead of merging into one

Happy to pick any of these up if you'd rather have them in scope here instead of as follow-ups.

Verification

mix test gives 347 tests, 0 failures, run just now against this branch at its current head, in a container matching .github/workflows/elixir.yml (Elixir 1.18.4 / OTP 28, postgres:15, same env). mix format --check-formatted and mix credo diff --from-git-merge-base origin/main are both clean. Earlier commit messages on this branch quoted different counts (336, 339, 341); those were snapshots from different points as tests were added, and 347 is the current, current-head number.

Both fixes above have tests that genuinely fail without them. Applying only the new test files to this branch, with no source fix, reproduces all three symptoms: submit-word against a live choice poll leaves ["some option 1", "some option 2", "injected"] on that poll, submit-word with a form on screen raises Ecto.ConstraintError on poll_opts_poll_id_fkey and terminates the LiveView, and the cloud renders its words while show_results is false. With the fixes in place all of them pass.

End to end against a real running instance: created a word cloud poll as the moderator (confirmed the options UI disappears when switching the type), enabled it, submitted a word as an anonymous attendee, watched it render as the cloud immediately. The size scaling matches word_size/3 on real numbers: with 4 attendees on one word and 1 on another, the 80% word renders at font-size: 41px and the 20% word at font-size: 21px.

On CI

For transparency, since this PR previously quoted test numbers that upstream CI never confirmed: the Elixir CI run on this PR from 2026-08-16 failed at "Check Credo Warnings" and its "Run tests" step was skipped, so the suite never ran upstream at all. 08425e1 addressed those Credo findings. The run queued against the current head is sitting in action_required, GitHub's approval gate for workflow runs on pull requests from forks, so it needs a maintainer to approve it before it will execute. Every number above comes from my own container run, not from upstream CI.

Fixes #79