Turn-Based Strategy With a Competent AI Opponent

Build a turn-based territory strategy game as one self-contained HTML file (inline JS, no libraries). Game: [describe rules precisely, e.g. "7x7 Ataxx-style board: click one of your cells to clone to an adjacent empty cell, or jump 2 cells (the source cell is emptied); all enemy neighbors of the destination convert to your color; board fills = game over; more cells wins"].

The AI opponent is the critical requirement:
- Implement opponent as a heuristic scoring function over every legal move: prefer moves that capture the most enemy cells, that increase total mobility (adjacent empty cells), and that avoid leaving a cell vulnerable to immediate recapture. Weight captures > mobility > safety.
- Evaluate ALL legal moves every turn (board is small); pick the best with a small random tie-break so games differ.
- Add a difficulty switch: Easy = random legal move, Normal = heuristic, Hard = heuristic + 1-ply lookahead that maximizes (my gain - your best reply gain).
- Move times must be under 100ms; no blocking the UI thread noticeably.

UI: click/tap to move; last-move highlight; simple animated capture flips; score bar; win/lose overlay with restart; difficulty selector on the start screen. 960x540 letterboxed, responsive, touch-first.

After the code, print the opponent evaluation function and explain each weight in one line.