SVP & CVP Improvements

[malb]

• SVP.shortest_vector now mostly does the right thing, i.e. it is no longer the wrong thing to call
• both bomb out with a helpful error message when the dimensions are too large.

[lducas]

Hi Martin,

I would suggest to first use follow a pre-computed strategy, even if it has small success proba. This provides a reasonable approx of the shortest vector, and you will get much better pruning coefficient for a second run targeting 99% success proba. Furthermore, this also provide a good guess at the optimal pre-processing choice, relieving the user from the burden of a delicate choice.

[malb]

Fair enough, might as well go all the way.

[malb]

This should be done now.