Fix d min issue

Resolves #223 (closed), #204 (closed)

When predicting peaks using a d min of lambda/2 and a minimum neighbour count of zero, the peaks now cover the entire detector scene, as they should.

In Qs2Events, the bisection algorithm to compute the frame number (sample angle) at the Ewald sphere intersection was flawed, assuming that the sphere is crossed exactly once. It's hard to know exactly how many times the sphere should be crossed, or whether it varies from system to system, but I found that dividing the frame range into 4 intervals and doing a bisection search on each found the missing events. Using 8 intervals resulted in no improvements over 4.

Note that in order to get all peaks, it's necessary to set the minimum number of neighbours to zero in the GUI. It's not clear whether this is the desired behaviour for the user.