Add more rigorous testing for exact optimizations

[sroelants]

There's a whole class of optimizations where we know they should not change the search result, even if they might change the returned score.

• Transposition table lookups
• Move ordering (MVV, killer moves, etc...)

A good way to test these features, then, is have a test suite of positions to run over, and make sure they always return the same move as when the optimization is disabled.

Note that this kind of stuff doesn't work for non-exact optimizations like null-move pruning or
quiesence search, because those will inherently change the resulting move.