ref(grouping): More parameterization metric improvements

[lobsterkatie]

This is a follow-up to #110612, making a few more improvements to our message parameterization metrics.

• Until recently, message parameterization was only ever the result of a call to normalize_message_for_grouping, so timing that function was a good proxy for timing parameterization itself. However, in ref(grouping): Cache parameterization on context #110605 we changed it so that a) normalize_message_for_grouping no longer calls the parameterizer every time it runs (instead relying on a cached value), and b) the initial calculation of that cached value happens elsewhere. normalize_message_for_grouping still can call the parameterizer, but it now only does so in cases where it has to parameterize something other than the event's primary message (in the case of error chains, for example). Thus timing it is no longer a good way to measure how fast parameterization runs. This fixes that by adding a timer to the parameterization itself.

• With that timing metric in place, we no longer need the parameterizer_called metric, because we can just use the timing metrics .count values to get the same information, so this removes it in favor of adding the experimental and changed tags to the timer.

• Right now, parameterization can be tagged as being experimental even if the experimental parameterizer isn't actually any different than the default one. This fixes that by checking the two regex sets against each other before setting the parameterizer's _experimental attribute.