Micro-optimize `normalize_stacktraces_for_grouping`

[Swatinem]

The function is slow, presumably because apply_modifications_to_frame of grouping enhancers is slow.
Here are a couple of ideas of how to optimize that.

Consider the following example enhancement rule:

family:native function:_CxxThrowException                         ^-group -group ^-app -app

This rule has 4 actions.

sentry/src/sentry/grouping/enhancer/__init__.py

Lines 128 to 132 in ad474a7

	for rule in self._modifier_rules:
	for idx, action in rule.get_matching_frame_actions(
	match_frames, platform, exception_data, cache
	):
	action.apply_modifications_to_frame(frames, match_frames, idx, rule=rule)

As the rule itself is a modifier_rule, all of its 4 actions are being applied:

sentry/src/sentry/grouping/enhancer/__init__.py

Lines 335 to 336 in ad474a7

	for action in self.actions:
	rv.append((idx, action))

Even though the group action does not do anything in this specific case:

sentry/src/sentry/grouping/enhancer/actions.py

Lines 93 to 95 in 4b7fc05

	# Grouping is not stored on the frame
	if self.key == "group":
	return

Idea: Further split rules/actions when the modifier/updater split happens to useless actions are not called.

Then again, ^-group -group is two separate rules that are applied separately.

Idea: Maybe we can merge these two rules into one? We would then have a rule with an inclusive range, instead of two rules with an exclusive range and an exact index.

[Swatinem]

Splitting of enhancement rules was implement in #51358, and #51781 was opened as a followup to track ideas about how to merge multiple enhancement rules into one.