When cohort counts are 0 warning not error should be thrown

[azimov]

When a cohort table is empty CohortDiagnostics assume that CohortGenerator has not been ran; this is not necessarily the case as the cohorts may be empty. When Running on multiple data sets this creates problems as the execution will not finish cleanly and an error is thrown.

[gowthamrao]

Also related to this #754

as why not run orphan concepts and other diagnostics that only rely on cohort definition expression and not instantiated cohort table