feature[next]: Add power unrolling functionality and respective unit tests.

[SF-N]

PR Title: Add power unrolling functionality and respective unit tests.

<type>:
    - powers of SymRefs and FunCalls until order 5 are unrolled per default
    - powers with higher exponent are still computed by the built-in functions.
    - test: add respective tests

<scope>: next 

[tehrengruber]

Suggested change

	if int(new_node.args[1].value) > self.max_unroll:
	base, exponent = new_node.args[0], int(new_node.args[1].value)
	if exponent > self.max_unroll:

and so on

[tehrengruber]

Suggested change

	return (
	"""Matches expressions of the form `power(base, integral_literal)`."""
	return (

[tehrengruber]

Suggested change

	def _compute_integer_power_of_two(exp):
	def _compute_integer_power_of_two(exp: itir.Expr) -> int:

[tehrengruber]

My bad:

Suggested change

	def _compute_integer_power_of_two(exp):
	def _compute_integer_power_of_two(exp: int) -> int:

[tehrengruber]

Suggested change

	def apply(cls, node: ir.Node, max_unroll=5) -> ir.Node:
	def apply(cls, node: ir.Node, max_unroll: int=5) -> ir.Node:

[tehrengruber]

Suggested change

	# In case of FunCall: build nested expression to avoid multiple evaluations of base
	# nest target expression to avoid multiple redundant evaluations

For clarification: It is not only that we don't want to reevaluate the base expression, but also the various power compontents themselves.

[tehrengruber]

Suggested change

	return InlineLambdas.apply(ret)
	return InlineLambdas.apply(ret, opcount_preserving=True)

The proper way would be to somehow tell the inliner to not ignore anything of the base, but we don't have a mechanism for this and the alternatives are less readable. This should be fine.

[tehrengruber]

Suggested change

	pytest.xfail("Not implemented.")
	pytest.xfail("Not implemented as we don't have an easy way to determine the type of the one literal (type inference is to expensive).")