Perf/avoid redundant fetches

[dbuezas]

Plotting multiple attributes or statistics of the same entity resulted in duplicated requests to home assistant.
This problem will be exacerbated with the inclusion of offsets.

This PR addresses this by:

• Caching the raw fetched data before picking an attribute (e.g current_temperature, target_temperature,...) or statistic (min, max, ...)
• Making multiple requests for the same entity in sequence (currently superfluous, but relevant for offsets), so the ranges subtraction logic avoids fetching overlapping intervals
• Delaying the decision of what attribute/statistic to pick from until plotting time

The performance improvement is exactly proportional to the amount of traces on the same entity. E.g, when plotting max, min and mean, it is 3x faster