Full date and round in game lists; engine evals from White's side

[jozef2svrcek]

Summary

The Players page (and the Games page, which shares the list) showed only the year, and games played on the same day came back in no particular order.

• Date column replaces Year. It shows the known part: 2026-03-15, 2026-03 or 2026.
• Round column is new. It's blank when the round is unknown (? or -).
• Order: newest date first, then latest round first, so a tournament's games read in playing order. Rounds are compared as numbers on their round and sub-round parts, so 10 sorts before 9 and 3.2 before 3.1. Game id breaks any remaining ties, so infinite-scroll paging never skips or repeats a game.
• Partial dates (2026-??-??) now sort below that year's fully dated games. Before, they sorted above them, because ? sorts after digits.
• Saved column widths are merged with the new defaults, so existing layouts keep working.

Also in this PR: engine panel and Analysis rail

• chessdb evals are shown from White's side, like Lichess. chessdb scores moves for the side to move, so with Black to move a good move was positive there and negative in Lichess. Now negative always means Black is better.
• Lichess evals are colour-coded like chessdb's. In both lists green/red means good/bad for the player to move, matching the !/? marks, not the displayed sign.
• Scrollbar no longer covers the close ✕ in the Analysis page's column of open games. WebKitGTK draws the scrollbar over the content, so the column now has right padding.

Cost

With no filter, the Games page sorts every game. On a synthetic 14M-row table each page takes about 125 ms instead of about 6 ms: the round expressions are computed for every row. A player's list is a few hundred games, so the change is negligible there.

Test plan

• cargo test -p chess-db: 107 passed
• Client typechecks and vite build succeeds
• Test database with same-day games in rounds 7–11, 3.1/3.2 and a year-only date: order is 03-15 R11 → 03-14 R10, R9 → 03-13 R8, R7, 3.2, 3.1 → 2026-??-?? → 2025. The offset page continues correctly.
• Checked in the app on the real database (drafts 0.17.2 and 0.17.3)

🤖 Generated with Claude Code

https://claude.ai/code/session_01FzbThuPFuAz9VYWoZd4ESu