Use sigma_T from astropy constants instead of a hardcoded value

[ftsamis]

Astropy now has a constant defined for sigma_thomson. This PR replaces the hardcoded value with it.

Props to @chvogl for discovering this.

For tardis_example outputs before and after this patch see: https://gist.github.com/ftsamis/eccbcc1b2a29053a412a2d74e5a11e7f

[yeganer]

Please update the reference date files.

[ftsamis]

@yeganer Not sure what you are referring to. Do you mean the test_cmontecarlo module? If so, done.

[yeganer]

No, I mean test_data.h5 and the numpy reference data files for test_tardis_full.py

[wkerzendorf]

@ftsamis @chvogl I like this change. Can we get this into a mergeable state?

[chvogl]

@wkerzendorf As @yeganer pointed out someone would have to update the reference data files (test_data.h5 and the numpy reference data files for test_tardis_full.py) before we can merge this.

[unoebauer]

I'd still like to merge this. Let's resolve the conflicts and update the reference data.

[unoebauer]

I am a bit puzzled why some of the test differ by so much since the change in the value of the Thomson cross section is really minuscule:

Before we used the hard-coded value of

6.652486e-25 

Now, we rely on astropy.constants, which includes a value for the Thomson cross section of

6.652458734e-25

in version 1.3.3.

This is a change on the level of 1e-6. Yet, in some tests, the differences to the reference values are on the per cent level.

Any thoughts on this, @wkerzendorf, @chvogl?

[unoebauer]

Closing. Work continues in PR #863