feat[next]: Extend and refactor constant folding

[SF-N]

This replaces this PR.
It refactors the ConstantFolding pass doing transformations until a fixed point is reached, like in the CollapseTuple pass. The FixedPointTransformation class from PR 1826 is used here.

The following new transformations are introduced:
CANONICALIZE_OP_FUNCALL_SYMREF_LITERAL:
literal, symref -> symref, literal, prerequisite for FOLD_FUNCALL_LITERAL, FOLD_NEUTRAL_OP
literal, funcall -> funcall, literal, prerequisite for FOLD_FUNCALL_LITERAL, FOLD_NEUTRAL_OP
funcall, op -> op, funcall for s[0] + (s[0] + 1), prerequisite for FOLD_MIN_MAX_PLUS
CANONICALIZE_MINUS: a - b -> a + (-b), prerequisite for FOLD_MIN_MAX_PLUS
CANONICALIZE_MIN_MAX: maximum(a, maximum(...)) -> maximum(maximum(...), a), prerequisite for
FOLD_FUNCALL_LITERAL: (a + 1) + 1 -> a + (1 + 1)
FOLD_MIN_MAX:
maximum(maximum(a, 1), a) -> maximum(a, 1)
maximum(maximum(a, 1), 1) -> maximum(a, 1)
FOLD_MIN_MAX_PLUS: maximum(plus(a, 1), a) -> plus(a, 1)
maximum(plus(a, 1), plus(a, -1)) -> plus(a, maximum(1, -1))
FOLD_NEUTRAL_OP: a + 0 -> a, a * 1 -> a
additionally to the existing ones:
FOLD_ARITHMETIC_BUILTINS: 1 + 1 -> 2
FOLD_MIN_MAX_LITERALS: minimum(a, a) -> a
FOLD_IF: if_(True, true_branch, false_branch) -> true_branch

In the end, unary minuses are transformed back to minus-calls where possible by UndoCanonicalizeMinus: a + (-b) -> a - b , -a + b -> b - a, -a + (-b) -> -a - b

Previously, large nested maximum expressions like maximum(maximum(maximum(maximum(sym, 1),...), 1), maximum(maximum(sym, 1), 1)) caused timeouts in PMAP-GO.