When calculating space usage for new fields in time-series inserts we don't take into account that fields are repeated in time-series buckets due to the control min/max fields.
https://github.com/mongodb/mongo/blob/2d46c87afcdf34a384d10e37b7b9b6ca2986fdf4/src/mongo/db/timeseries/bucket_catalog.cpp#L316-L322

This seems to be causing differences in how many measurements we can store in a single bucket depending on which batch size is used during insert.