Implement #87: TSDFRecording based API

[twanvl]

This PR implements a more convenient interface for working with single-recording tsdf files.

See #87 for a detailed description and motivation.

The most basic use case now becomes:

df = read_dataframe(input_file)
write_dataframe(output_file, df)

[twanvl]

The new functions should probably also be exported from tsdf/__init__.py

[twanvl]

This also warrants an increase in the package version number.

[Erikpostt]

Hey @twanvl, overall this looks very neat -- I'd like to merge this soon. I do have a few questions:

• What is the reason you suggest to use "recordings"? Wouldn't it be consistent to use "time_series"?
• You suggest using read_single_recording_metadata, read_multi_recording_metadata, and read_single_recording_dataframe. What do you think about combining the first two into read_recording_metadata, where: if len(recordings) ==1: return recordings[0] / elif len(recordings) > 1: return recordings / else: raise Exception()
• If I load a dataframe using tsdf.read_single_recording_dataframe, and specify the channels parameter to a subset, it might be nice to automatically change the channels parameter of the .attrs attribute of the dataframe. Otherwise, I get AssertionError: The channel groups do not match the columns in the DataFrame when using tsdf.write_dataframe().
• Perhaps you could add the functions the user wants to use (I think the reading/writing functions) to the __init__.py.

Let me know what you think. We could also merge your suggestions, and add issues related to my questions if you think they are valid.

[twanvl]

• The name "recording" was used in some tsdf documentation I believe. If can also see the argument for TimeSeries or Series, and I am fine with changing it.
• That is what read_single_recording_metadata does. But there can be meta data files that contain multiple recordings/time series.
• to automatically change the channels when loading a subset: Yes, that makes sense.
• Adding to __init__.py: Yes that should be done.

[Erikpostt]

@twanvl you want me to implement these suggestions to your PR?

[twanvl]

I have fixed all the issues mentioned above.

I also added an add_channel function to TimeSeriesMetadata, to make it easy to actually add channels and not just remove them. See the tests for an example.

I also added tests to ensure that all non-standard metadata is preserved when reading/writing.