Limited by std::vector creation, deallocation and copying

[asalzburger]

Here's something that might be of interest @krasznaa @stephenswat @paulgessinger @niermann999.

I have now implemented a system for a single layer test, one barrel layer, one end cap layer with the following tests:

• a barrel layer has N surfaces and an approach surface: I shoot M particles onto the approach surface and then ask a local grid for the indices of all transforms of surfaces in question, and then intersect only those, this is usually 6 or 9 surfaces per test

• then I do the same test with a trial-and-error, i.e. testing all surfaces for intersection running indices from 0 to N and take only the intersected ones, this is 12 x 7 = 84 surfaces per test

Good thing: both methods achieve the same result, below are the pictures for barrel and end cap case:

Now, here's the interesting part, the second version is actually faster in my tests:

Because half of the time is actually spent in finding out and especially allocating the return vector of indices to the program:

I.e. the O(ns) intersection estimation for the surfaces outperforms the data storage model ...

[asalzburger]

I could speed up the grid::zone(...) access for cases where the bare_value is equal to the store_value (i.e. grids with single entry per bin) by avoiding insert and bin_content creation.

Something that should work for all fixed size store_value grids.

[asalzburger]

This is in #38