Design for multivariate series

[kylebrandt]

Currently the mathexp.Series type of data.Frame only support a series with a single time column and a single value column. To represent (label/tag) like responses you would get from OpenTSDB/Prometheus that tags are a Property on the Frame. (For consideration is that it could maybe be a property on Fields (columns) instead.

A multivariate series still have a single time column, but multiple value columns (see https://docs.google.com/document/d/1l35soEUiUOMiBSQTXxisxLSz1E5vh9DFYw9xgxy5ocM/edit#bookmark=id.ecr5mfkb49j7). This would be a common response from SQL like data sources and Influx.

Frontend Transformers have some support from this from what I saw from @dprokop's demo the other day, so we should coordinate there.

[kylebrandt]

The main area of confusion is how we will be working with different data models. In particular the tag/label model vs a tabular model. The TimeScale Data Models docs describe these as the "Narrow-Table Model" and "Wide Table Model".

Ryan and I took some brainstorming notes in google docs but don't have a clear next step yet.