ADPs in structure factor calculation, continued

[apeck12]

Thanks for the feedback! Benchmark (1000 q-vectors, 1000 atoms, 1000 rotations) for the original cpuscatter function prior to including ADPs:

$ time ./cputest
CPP OUTPUT:
0.000000
0.000000

real 0m11.691s
user 0m11.682s
sys 0m0.008s

For the original ADP implementation, it clocked in at:

$ time ./cputest
CPP OUTPUT:
0.000000
0.000000

real 0m18.371s
user 0m18.366s
sys 0m0.004s

I revised this function so that Debye-Waller factors are pre-computed in the first nested loop (which loops over q-vectors) rather than in the third nested loop. However, the improvement in speed is marginal:

$ time ./cputest
CPP OUTPUT:
0.000000
0.000000

real 0m16.515s
user 0m16.511s
sys 0m0.003s

I only revised the CPU code, as a comment in cpp_scatter.cu indicates that caching pre-computed Debye-Waller factors could pose memory problems for the GPU version.