Split DateTime symbols data into length-based chunks

[Manishearth]

Part of https://github.com/unicode-org/icu4x/issues/3413

Currently DateTimeSymbols data contains data for every length of symbol data, regardless of which is requested. We shoudl slice it down into MonthSymbols (which can be long/short/medium/full), WeekSymbols, etc and load just what is needed during data loading.

This is likely a 2.0 candidate (https://github.com/unicode-org/icu4x/issues/2856). This will need some investigation.