Disable refinement hole fits

[pepeiborra]

Let's review the benchmark results to see how expensive refinement hole fits are in practice. Then we can decide whether to keep them or not

[pepeiborra]

Results for the lsp-types example:

upstream   edit                             True      50        1.671270325          0.0           0.6073194759999999    8.608297202000005       9.223011378          202MB          9026MB
HEAD       edit                             True      50        1.305338329          0.0           0.38494863600000007   10.012330334999998      10.402405344         203MB          9072MB
upstream   hole fit suggestions             True      50        1.142132948          0.0           189.604739854         1.9936279e-2            189.630975323        282MB          199790MB
HEAD       hole fit suggestions             True      50        1.890731358          0.0           85.15542919800005     1.9193362999999998e-2   85.181236653         268MB          85096MB

Disabling refinement hole fits improves the total time by 50%. It still takes 9X longer to typecheck than without any holes at all, but I guess this is progress.

@anka-213 what do you think about not enabling refinement holes? Users can always enable them in their cradles.

[Tritlo]

Just a comment: typed-holes (and refinement holes) will become a lot faster soon https://gitlab.haskell.org/ghc/ghc/-/merge_requests/5980