Constant folding flags are misleading

Two problems

• The flag -fcore-constant-folding claims to enable/disable constant folding. But it doesn't. The only use is here in GHC.Core.Opt.Pipeline.getCoreToDo:

    const_fold    = gopt Opt_CoreConstantFolding          dflags
  ...
    do_simpl3      = const_fold || rules_on -- TODO: any other optimizations benefit from three-phase simplification

  That do_simpl3 switches on the big 3-stage application of the optimiser.

  There is no selective enable/disable for constant folding rewrites, independent of other simplifications.

• The flag -fnum-constant-folding is entirely un-documented. We have this commit:

  commit 60e4bb4d305bc1a65457ee79b1e69c11b9ed747d
  Author: Sylvain Henry <hsyl20@gmail.com>
  Date:   Fri Apr 13 13:29:07 2018 -0400
  
    Enhanced constant folding
    This patch uses laws of +/-/* (associativity, commutativity,
    distributivity) to support some constant folding into nested
    expressions.

  But the user manual is silent; and the code is silent too. No Note [Enhanced constant folding] to explain what the "enhanced" rules do, or how. Alas.