Replace scalar instances of np.sqrt, np.cos, etc. with math alternatives

[janbridley]

Description

Calls to numpy square root and trig functions that operate on a single value can be replaced with the corresponding math functions.

Motivation and Context

While the performance difference is small in absolute terms, the relative improvement is on the order of 10x. Changing these instances could yield a small performance bumps with only a little effort.

import numpy as np
import math

x = np.random.rand()  

# sqrt:
%timeit -r 100 -n 1000000 np.sqrt(x)  
> 398 ns ± 14.3 ns per loop (mean ± std. dev. of 100 runs, 1,000,000 loops each)

%timeit -r 100 -n 1000000 math.sqrt(x) 
> 24.3 ns ± 2.24 ns per loop (mean ± std. dev. of 100 runs, 1,000,000 loops each) 

# cos:
%timeit -r 100 -n 1000000 np.cos(x)                                                                                                              
> 395 ns ± 10.1 ns per loop (mean ± std. dev. of 100 runs, 1,000,000 loops each)  

%timeit -r 100 -n 1000000 math.cos(x)                                                                                                            
> 31.3 ns ± 2.55 ns per loop (mean ± std. dev. of 100 runs, 1,000,000 loops each)  

# arcsin:
%timeit -r 100 -n 1000000 np.arcsin(x)                                                                                                           
> 413 ns ± 17.5 ns per loop (mean ± std. dev. of 100 runs, 1,000,000 loops each)   

%timeit -r 100 -n 1000000 math.asin(x)                                                                                                          
> 35.8 ns ± 2.85 ns per loop (mean ± std. dev. of 100 runs, 1,000,000 loops each)   

[tommy-waltmann]

This is a good thought, however freud modules are compiled with cython. Numpy has its own cython modules and a compiled backend, so its possible that these benchmarks will be different if they are done inside a module compiled with cython.

I would be hesistant to make any changes related to this issue before seeing a cython benchmark.