Fix _from_ema_to_offsets

[alexandermorgan]

This method works and is useful but should be renamed to _ema2measures to better represent what it does. This also reveals that some ema addresses are faulty, such as observations 4701, 4702, and 4710. It's very helpful that this reveals these issues, but perhaps problematic lines should be dropped by default. The problem lines are the ones where the new "end" column == "". This is caused by the ffill functionality not working as intended (seems to be operating on the wrong axis).

[RichardFreedman]

There are indeed broken EMA references in CRIM (partly the result of fussy JS and selection tool). Having a way to identify those would indeed be helpful. Fixing them is tedious, however. I agree that dropping them from DFs would be a reasonable option for some analysis tools.