Enhance `valuehead` search algorithm

[mooskagh]

Like #2175, the valuehead algorithm is minimal and can use some improvements.

Valuehead is a bit less popular than policyhead, and a bit harder to tweak, but here are ideas of improvements:

• Make it a fixed depth minimax (rather than depth-1).
  • It feels that the alpha-beta pruning could be possible to make work (with some speculative prediction) with NN batches, that's something to explore, too.
• Support for Syzygy tablebases
• Multipv mode or other enhancements of the display (maybe something like --verbose-move-stats, although how it's done in the new API world is not decided yet)
• Emulate "thinking", but having a delay before making the move (that got mentioned quite a few times in Discord)
• (currently valuehead is inherited from InstamoveSearch. It may or may not be a good idea to detach i)

It would be a good project for somebody who would like to familiarize with the Lc0 codebase, as it's touching some of the core Lc0 parts, without being on a "mainstream" Lc0 code path.