[ADD] Option to treat weight and bias of a layer jointly in KFAC

[f-dangel]

Eventually, this will resolve #53.

At the moment, I am struggling to get the tests running and would benefit from a second pair of eyes to look through the code. At the moment, the tests seem to execute for both separate_weight_and_bias = [False, True], but the matrix produced by KFAC is wrong for True.

I believe this is related to the order of parameters. But I think I am actually arranging them in the correct order 😂, so maybe I'm barking up the wrong tree. In any case, it would be great if you could skim the code and look for anything obvious I missed.

[f-dangel]

Update: I found the problem and fixed everything. Ready for review.

[coveralls]

Pull Request Test Coverage Report for Build 6801539850

• 22 of 22 (100.0%) changed or added relevant lines in 1 file are covered.
• 40 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-4.8%) to 87.668%

---

Files with Coverage Reduction	New Missed Lines	%
curvlinops/examples/utils.py	6	41.67%
curvlinops/fisher.py	34	28.85%

Totals
Change from base Build 6790452156:	-4.8%
Covered Lines:	718
Relevant Lines:	819

---

💛 - Coveralls

[runame]

Sorry was asleep :D But LGTM!