Add sample-access to read_dataframe

[matthiasstraka]

Currently, all access to data is through time in seconds. Internally, we already use sample counts. Provide a suitable interface for sample-wise access.