[ADD] Support KFAC if the network has inplace activations

[f-dangel]

Resolves #58.

[coveralls]

Pull Request Test Coverage Report for Build 6802882104

• 6 of 6 (100.0%) changed or added relevant lines in 1 file are covered.
• 40 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-4.7%) to 87.82%

---

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 6801631981:	-4.7%
Covered Lines:	721
Relevant Lines:	821

---

💛 - Coveralls

[runame]

Can't we also apply the same refactor as in SINGD, where we just use a single forward hook that computes the input covariance and installs the tensor hook on the module output?

[f-dangel]

One point against merging these hooks is that we may want to re-cycle the hook for computing the activation variance (e.g. FOOF).

[runame]

Ok let's stick to separate hooks for now then.