make sure all cohort manipulation functions work within observation periods

[edward-burn]

As well as #256, make sure that any other functions that collapse cohort entries respect observation periods. If one cohort entry is in one observation period and another is in a second observation period then we should not concatanate them as we will get a cohort entry that goes outside of observation.

We should use observation period id as a grouping variable.

[edward-burn]

functions to add tests to check this behaviour:

• collapseCohorts()
• intersectCohorts()
• unionCohorts()