fixing calculation of average

[grzanka]

This PR introduces couple of changes in how multiple runs are averaged:

• online version of mean and variance calculation methods are introduced and documented in averaging.py module
• error attribute of page is now initialized to None, instead of array of same shape as data - this reduces memory if no error calculation is requested
• tests are adjusted accordingly
• support for python 3.8 is dropped to fully benefit from type hinting syntax introduced in 3.9

[grzanka]

To test create a virtual environment and run:

pip install git+https://github.com/DataMedSci/pymchelper.git@722-check-if-weighted-average-from-multiple-runs-are-ok#egg=pymchelper

If working on HPC (Ares) remember to include new python ml python first, but do not load official pymchelper package via ml pymchelper.

[grzanka]

TODO:

• weights needs to be added to all aggregators update method
• weights needs to be seeded (as number of simulated primaries) here

  pymchelper/pymchelper/input_output.py

  Line 120 in d6c5ac3

  aggregator.update(page.data_raw)

[grzanka]

@nbassler that should be ready for review now

[nbassler]

maybe add the above comments for clarity?

(or maybe a table listing this incl stddev and stderr calculation)

[nbassler]

Checked this against rev49 by sha256sum for timepix data . Only difference found was png images, which have changed due to the /prim.

OK