Introducing a better return type for `fn percentiles(...)` results

[thinkingfish]

A few things we could improve:

• using a HashMap instead of Vec makes the results easier to use;
• renaming the function to be fn quantiles() which is a more accurate/correct description of the query;
• return the total count across all buckets, as it gives us a sense about the statistical meaningfulness of a specific quantile (e.g. the value of p99.9 from a histogram with 10 samples doesn't have the same statistical significance as p99.9 against a 1 million samples);
• return the Bucket for min (p0) and max (p100) to help contextualize the results for typical use cases;
• (requires changes to the Histogram struct) track under- / over-flow of values and return them with the rest of the result.