143 refactor notch approximation

[johannes-mueller]

Closes #143

In a nutshell:

• Introduce NotchApproxBinner new class to do the binning for the notch approximation.
• Introduced methods NotchApproximationLawBase.primary and NotchApproximationLawBase.secondary in order to be able to get stress strain data in only one lookup.
• Therefore the load argument can be dropped in the methods for strain calculation.
• NotchApproximationBinner has the method .initialize to set the maximum load. That means, that determining the maximum load can be left to FKMRainflowDetector.
• Drop obsolete Binned class in favor of NotchApproxBinner
• Adjust the unit tests accordingly. Some assertions in tests/strength/fkm_nonlinear/test_fkm_nonlinear.py had to be dropped because they were using internals of Binned.

This adds a performance boost of roughly factor 10.

Open point

As we now do no longer need to precondition the binner with the maximum load, we need to rethink the role of "max_load_independently_for_nodes" in the assessment parameters and find a neat way how to handle it. This will be treated in a follow up issue.

@maierbn take your time to review, no rush. We can discuss details after the winter break. I put some effort into arranging the commits in a meaningful way. So reviewing them individually might be easier to understand what's going on.