Percentiles - empty df handling and unit tests

I created this MR after running @shamshaw's example notebook and ran into some issues handling empty data frames from certain states. This MR adds some specific code that handles empty input dataframes and outputs empty percentile dataframes where all values are nans. I know that Scott created another pull request that alleviates the same issue, but I was running into errors running this approach line by line -- it's possible I missed something in the function. This approach is a bit more up-front about empty df handling. I also added some unit tests.

This is draft because I think we need to talk through how we want to handle this throughout the package. I am happy to go through and make handling consistent across the package.