Properly handle memout and timeout for quality optimization and memout for runtime optimization

[mfeurer]

PR #145 only fixes the bug that SMAC crashes if the first run is a timeout or a memout. If these occur more often, the location of these should be taken into account when suggesting new points. There are several solutions we could implement:

• https://arxiv.org/pdf/1403.5607.pdf
• https://arxiv.org/pdf/1004.4027v2.pdf
• https://pdfs.semanticscholar.org/2e07/9604c7a00c43f06e214280cea18a89dcecef.pdf

[mfeurer]

And a follow up issue which can only be solved once this is done: if we properly handle crashed runs, there is not always a reason to abort if the first run crashed. A crashing first run might occur with a random incumbent and a pipeline which has unknown constraints (for example an optimization algorithm in a ML method does not converge due to the dataset).

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[alexandertornede]

We will look into this together with #884.