split reading into init and separate read call

[jgriesfeller]

In order to get pyaerocom's issue metno/pyaerocom#1301 done, we need to split the reading into two calls:

• Init of the reader class
• do the actual reading

This is needed so that revisions can be checked e.g. for caching on pyaerocom's side.

[heikoklein]

Sorry, what pyaerocom calls caching is providing a separate unified fast reader class.

If a pyaro-reader is too slow, it should be replaced by a fast pyaro-reader, e.g. netcdf-rw. pyaro should not need caching.

[jgriesfeller]

This is also to avoid reading the data twice as it is done now. I don't think we should really read data via class init. Not sure yet how that will work...
On top we tend to implement filtering in the reading classes now (for speed). Not sure how all of that fits together yet.
To be discussed and to be seen.

[heikoklein]

I agree that there should not be any heavy reading during the init, and pyaro does not require that. It is up to the pyaro-reader to decide when and where it reads the data, and some might require to read all files to get any information.

[jgriesfeller]

Will not be implemented. just read on first data access