Add support for different strategies for doing percentile aggregation

[dblock]

Is your feature request related to a problem? Please describe.

Coming from #3634, OpenSearch percentile aggregation has produced different results across versions on estimation boundaries. Multiple results are correct, but users may be confused by these results because different implementations used different ways to perform quantile approximation and because other software used by developers (e.g. Julia or Google Sheets) produce different numbers than OpenSearch.

Describe the solution you'd like

• Global settings that let users perform quantile estimation similar/identical to results produced by R, Julia, or Google Sheets.
• Query options that let users perform quantile estimation similar/identical to results produced by R, Julia, or Google Sheets.

Additional context

We use t-digest for this, but some strategies could use a different, albeit slower, implementation as an option. Do we know of any?

See #3634 for an engaged discussion with the author of t-digest about what the "correct" values are, @tdunning.