Disable puzzles with too few alternative moves so they're no longer served

[Copilot]

Some puzzles start in check with only forced replies, letting players get stuck in a perpetual-check loop when they deviate from the recorded solution (e.g. 3oY3UwhN, ohT0nTBS). This adds a way to detect such puzzles, flag them, and exclude them from assignment.

Schema

• Liquibase changeset 0108 adds a nullable disabled_at timestamp to puzzle (a puzzle is disabled when disabled_at is set); regenerated the codegen changelog.

Detection

• New PuzzleSolvabilityValidator.hasEnoughMovesAtEachPlayerStep(...) (located in the scripts/puzzles package, not the service layer): replays setup + solution moves and requires ≥ MIN_LEGAL_MOVES_PER_STEP (3) legal moves at every player turn. The player is the side to move after setup; opponent turns are skipped.
• New DisablePuzzlesWithoutEnoughMoves script (in the scripts/puzzles package) iterates all puzzles and sets disabled_at = now() for those that fail the check.

Serving

• PuzzleCache keeps a boolean enabled per entry, derived from disabledAt == null. PuzzleCache.randomId now selects only from enabled entries. The cache still loads disabled puzzles, since they're needed for by-id lookups and played-puzzle history.

val enabledEntries = entries.filter { entry -> entry.enabled }
var filtered = enabledEntries
// ...apply exclude / rating range / category filters...
return if (filtered.isEmpty()) enabledEntries.random().puzzleId else filtered.random().puzzleId

UI

• The puzzle's enabled flag is now surfaced through PuzzleResponse and the client-side PuzzleDto. The puzzles page shows a warning icon next to the puzzle id when a disabled puzzle is loaded (e.g. via a by-id link), with a tooltip explaining it is no longer served.

Rating

• Playing a disabled puzzle no longer transfers any ELO: processOutcome checks the puzzle's enabled flag from the cache and applies a no-op transfer (ratings left unchanged) when the puzzle is disabled.

Notes

• If every puzzle were disabled, randomId throws NoSuchElementException, already handled upstream as "Puzzle cache not ready yet".
• Added a changelog entry.

[benckx]

@copilot drop the test

[Copilot]

Dropped the test in d5dd623.

[benckx]

@copilot resolve the merge conflicts in this pull request

[benckx]

@copilot Add a column in puzzle table for date of insertion. For existing puzzles, check in changelog.html for an appropriate date, that would be around 3 years ago.

[Copilot]

@copilot Add a column in puzzle table for date of insertion. For existing puzzles, check in changelog.html for an appropriate date, that would be around 3 years ago.

Added changeSet 0116 with a nullable created_at column on puzzle. Existing puzzles are backfilled with 2023-03-25 — the day the initial batch of ~18,000 puzzles was uploaded (per changelog.html). New puzzles created by CreatePuzzlesFromCsv set created_at = now(). Regenerated the codegen changelog.

[benckx]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Merged main and resolved the conflicts in 7fc3ba9:

• liquibase changelogs: main had taken changeSet id 0115 (content_section_vote.notified), so the puzzle changeSets were renumbered — disabled_at → 0116, created_at → 0117. Verified: liquibase applied cleanly through 0117.
• puzzles.css: kept main's #puzzle-feedback-info-box → #puzzle-state-info-box rename (and the mobile flex order rules) alongside this branch's .puzzle-disabled-icon styles.