ref(grouping): Improve message parameterization metrics

[lobsterkatie]

This makes a number of adjustments to our parameterization metrics, both to improve them and to reflect recent changes to the code.

• Right now, we have a message_parameterized metric, which only fires when the parameterized message doesn't match the original message, not every time we run parameterization. We therefore can't answer the question of how often parameterization has any effect. To answer that question, this adds a new metric inside the parameterize function, parameterizer_called, which fires any time the function runs, and which includes a boolean changed tag.

• Now that we cache parameterization results, firing the message_parameterized metric any time normalize_message_for_grouping is called isn't actually accurate anymore. This therefore replaces it with a new message_used metric, so that we can continue to track our various reasons for using messages.

• Caching parameterization results doesn't actually make things any better if we only use the cached value once. To better track how effective caching is, this adds to the context usage flags for both the parameterization result and the parameterizer itself (which is also cached, to save us having to check which one to use over and over). These are then checked (and flipped to True if they weren't already) each time we use one, and new metrics (cached_param_result_used and cached_parameterizer_used, respectively) are fired any time the flag indicates multiple usages.

[shashjar]

left a couple of suggestions