Add deliberate practice for reviewed mistakes

[sanderland]

The app gives feedback, but it does not yet turn a mistake into a later successful decision.

Add a small practice loop:

• Return to the position before a selected mistake.
• Hide policy and explored-continuation cues for an unaided retry.
• Let the player play a variation, then compare it with the original move and target-policy candidates.
• Allow bookmarking a decision and revisit it after a delay.

Keep the current quiet defaults. This should be an explicit review activity, not more automatic advice during normal play.