Fix score_entry function

[jarad]

The current version of the score_entry is simple and works if the scores did not include any additional bins. Due to the guidelines, these additional requirements need to be met

• For week targets, include the preceding and proceeding weeks. For example, if the truth is week 1, then this means including week 52 (53 in some years) as well as week 2.
• For percent targets, include the preceding 5 and proceeding 5 percents.
• Deal with the fact that the "Season onset" can possibly be "none".

[craigjmcgowan]

Rather than try to expand the truth, perhaps it would be easier to have a function that expands our definition of the "correct" cells on the entry data frame after joining entry with truth. This is what we did in the past and the weeks are already in the correct order on the submission.

We also need a way to take into account the weekly submissions of the 1-4 wk ahead forecasts that will have different truths for each submission.

[craigjmcgowan]

Perhaps it would be easier to index submissions by MMWR week they are based on, rather than submission date. That way if submission date changes due to holidays any functions will be unaffected, as the forecasts will still be based on the same MMWR week.

[craigjmcgowan]

We also need to build in capacity for multiple peak weeks - this happened this past year in HHS Region 8 where both weeks 8 and 11 had the same peak ILINet values.

[craigjmcgowan]

I believe I made score_entry fully functional and also completed the expand_truth function. expand_truth is a little slow due to the loops but when I ran it on a full season's worth of observed data it still only took 35 seconds. I still need to write tests for these - I will probably add some "correct" data frames to the data folder to use in those tests.