Fix numIncrease, pctIncrease, rolling calculations where dates are missing

[flaneuse]

See outbreak.info/epidemiology?location=ESP&log=false&variable=confirmed_numIncrease&xVariable=date&fixedY=false

1. For Spain, the two bad values are 4/28 and 5/2 and by my count should be 957.29 and 449.57, instead of -904 and -1828.
   data for the 4/28 value:
   4/25: 3995
   4/26: NA
   4/27: NA
   4/28: 2706
   4/29: NA
   4/30: NA
   5/1: NA

2. pctIncreases are wrong if a day is skipped. should be infinite % increase, not relative to the previous non-zero day

3. Consider adding in explicit confirmed=0, dead=0 when values are NA.

[flaneuse]

temp fix from @gkarthik : exlcuded negative numIncrease values from rolling calculations