Linking Across Data Granularity: Fitting Multivariate Hawkes Processes to Partially Interval-Censored Data