Statistics modules should write to time series DB

[cornelinux]

Statistics data are currently written to the database table monitoringstats. This table can get filled up quite quickly.

1. We need a way to rotate this data

2. We should allow stats modules to write to a separated time series database, that can handle this kind of data in a better way.

[cornelinux]

There are several time series databases around.
Maybe we should add an abstractin layer like the audit module and allow adding other database backends for the statistics.

The first possible module could be the SQLStats.
The module interface could look rather similar to the AuditModule.

We keep the library functions in monitoringstats.py like write_stats, delete_stats, get_stats_keys... but are using the corresponding new MonitoringModule.

After we made this modular, we could look for more preformant time series databases like Prometheus, Wisper or timescale DB.

Finally we should drop the old statistics (lib/stats.py) which is based on the audit log an uses pandas!

[plettich]

BTW: the tests for lib/stats.py didn't really work since no image was generated.

[cornelinux]

@fredreichbier To be able to write to different time series database we need the app-configuration in some task modules like in the do e.g. of the event counter task module:
https://github.com/privacyidea/privacyidea/blob/master/privacyidea/lib/task/eventcounter.py#L73

First I thought of passing the app-config in the do-method, but then I thought it might be nicer to pass the app-config in the constructor of the class.

But now I end up with problems passing a parameter to a constructer of the task modules in your nice lamda voodoo of TASK_MODULES...

e.g. here:
https://github.com/privacyidea/privacyidea/blob/master/privacyidea/lib/periodictask.py#L59

or here:
https://github.com/privacyidea/privacyidea/blob/master/privacyidea/lib/periodictask.py#L47

It might be also only a problem due to the Mock in the tests. Let's talk.