feat(plugin-multiplayer-vote): add anonymous group-vote primitive

[Mandyx22]

Summary

Adds plugin-multiplayer-vote — an anonymous group poll, the last of the Tier-1/2 multiplayer primitives (after live-scoreboard #36, choice #37, match #38, turn #39).

Every participant votes for one of the same options; the trial pushes that vote and barriers until all expected_players have voted, then optionally reveals the aggregate tally and the plurality winner.

How it differs from plugin-multiplayer-choice

• Aggregate, not individual — the emphasis is per-option counts + a winner, not who chose what.
• Anonymous by construction — the data and reveal carry per-option counts (tally/winner/tied_options), never a participant→vote mapping. The tally is labelled from the trial's own trusted choices, so no untrusted per-voter string is ever exposed. Each client still records/sees only its own pick (vote/vote_index + a "(you)" reveal highlight).
• Plurality winner — most votes wins; a shared top count is a tie (winner: null, is_tie: true, tied_options lists the leaders); zero votes → winner: null, not a tie.

Design

Barrier-then-reveal trial on the multiplayer API (a leaner choice, minus payoff/player_label):

• Pure core (vote-core.ts, jsPsych-free, exposed as statics): readVote, countVoted, tally, plurality.
• The barrier counts only in-range votes (countVoted(g, dataKey, choices.length)) so it always agrees with the tally — a stale/out-of-range index (e.g. left under a reused data_key) neither lifts the barrier early nor is dropped from n_votes.
• Supports a wait timeout (proceeds partial, flags timed_out), reveal: false, and custom button_html.

Tests

• vote-core.spec.ts + index.spec.ts — 34 tests (unit + MockApi barrier flow + startTimeline smoke test), including regression tests for the in-range barrier count.
• Repo-wide: 321 tests / 14 suites pass.

Review

Went through three high-effort /code-review passes; 6 findings surfaced and fixed (barrier/tally consistency, anonymity doc precision, tie-break example robustness, tie-state styling), final pass clean.

🤖 Generated with Claude Code

[htsukamoto5]

Pushed a robustness fix for the latest jsPsych#3694. The barrier's catch treated any api.wait() rejection as a timeout (timed_out: true + on_timeout), but #3694 now rejects a real timeout with a typed MultiplayerTimeoutError — and a wait() can also reject because the condition predicate throws or the backend fails. Those non-timeout faults were being silently mislabelled as a timeout.

Changes (commit on this branch):

• The catch now rethrows unless error.name === "MultiplayerTimeoutError". A genuine timeout still degrades to a partial tally; any other rejection halts the trial loudly. (Matching on error.name rather than instanceof survives two loaded copies of jspsych.)
• Set the test mock's timeout rejection to name = "MultiplayerTimeoutError" so it mirrors the real API, and added a regression test that a non-timeout wait() rejection propagates instead of being masked.

The existing push-failure test already covered the pre-wait path; this extends the same guarantee to the wait itself. tsc clean, tests + build green, patch changeset added. Same pattern as plugin-multiplayer-sync/-role (#45).

[jodeleeuw]

Closing in favor of #37: side-by-side review showed vote shared choice's entire choose → push → barrier → reveal engine and added no unique parameters, so its functionality has been folded into plugin-multiplayer-choice as reveal_mode: "tally" + record_choices_by_player: false (commit 6d0d449 on #37). The tally reveal, tally()/plurality() core, bounded barrier count, and tests were ported essentially verbatim — thanks @Mandyx22, that work all ships, just under one package. See the update comment on #37 for details.