Update MAE metrics to limit potential forecast deviation by up to 48 hours

[aaTman]

Currently, the method in which MaximumMAE is calculated takes the difference between the maximum temperature of the forecast at each unique instantiated forecast at each init_time (i.e. a 2023-01-01 00z forecast valid from 2023-01-01 00z to 2023-01-11 00z), and the maximum of the case.

There is a risk of edge cases in which a model captured - correctly, or incorrectly - an anomalous temperature event far enough in a forecast that it would be independent of the regime that we want to focus on for that case.

We approached filtering the events with a 48 hour buffer on the ends of each heat wave event with the intent of capturing onset, duration, and temperature deviations in case a model has a temporal displacement. With that in mind, we want to truncate the maximum temperatures in the forecast to +-48 hours of the occurrence of the actual maximum of the regionally averaged temperature based on ERA5 data.

I will submit a PR shortly with updated testing, but happy to invoke a discussion if anyone watching the repo has thoughts!