Fix problems with dates part 1

[trzysiek]

Query from the test caused Quesma to crash without any error message, as it'd run of out memory in 0,1s or so. You can see it below:


It was a small bug when adding empty rows. Usually keys in date_histogram are x, x+1, x+2,..., when we're using timestamp / duration as a key, so number of added empty rows was kind of limited. But in calendar_interval for big intervals, like 1 week, keys are timestamps in millisecond, so we were adding a new row for each millisecond, exhausting all the memory very quickly.
Fixed here by considering the second option.

[jakozaur]

@trzysiek perhaps we should add safeguard to not add more than X missing timeslots (like 1000). It looks like we still may hit this issue.