Confirm before closing a meeting (#229)

[aleksejs1]

Closes #229.

Problem

An employee with a filled-in draft scrolled to the bottom of the page for Publish and pressed Archive instead. It closed the meeting for both sides at once, with no question and no way back, and the draft could no longer be published.

What changes

Frontend only: no endpoint, schema or encryption change.

• Closing is two presses. The archive form's button and the "not closed" card's "Didn't happen" open an inline confirmation (anketa/ArchiveConfirm.svelte) and send nothing. Focus lands on Cancel; a double click or a held Enter only opens it.
• Never over my own unpublished answers. With answers in an unpublished draft, the only way to close is "Publish and close": closeMeeting() publishes first and archives only if that worked.
• Everything else is allowed, with a warning naming whose answers the closed meeting will lack — so a meeting the counterpart never published on can still be closed. The rule is archiveConfirmation() in anketa/archiveConfirmation.ts.
• The archive button is secondary and reads "Close and schedule the next one" / "Close this 1:1". Publish is the only primary button on an unpublished side.
• A cleared next-meeting date stops the close before anything is published.
• A publish refused with 409 (already published elsewhere, or archived) says to copy the text and reload, instead of the bare "Already published.".

Ten new strings in all 6 locales (written by Claude, not native-speaker-reviewed); anketa.archive removed.

Not covered: the published side closing while the counterpart is still writing. That needs an undo — #230.

See docs/decisions/2026-10-07-confirm-before-archiving.md.

Verification

• Unit tests for archiveConfirmation() (every combination) and hasAnswer(); 943 unit tests pass.
• Typecheck, lint, knip, format, duplication and doc-link checks clean.
• Full Playwright suite (68) against the e2e stack, with three new dual-actor tests: a draft is published before closing and the counterpart reads it; a tab that missed its own publish is told to reload and closes afterwards; the published side closes a meeting the counterpart never published on.
• Four code-review rounds, the last one clean.
• The two generation scripts (generate-demo-fixture.mjs, generate-doc-screenshots.mjs) were updated for the two-step close but not run.

🤖 Generated with Claude Code