Improve performance for large arguments in besselj0

[heltonmc]

The bug fix in #57 was fine for accuracy but this should be a lot better....

# fine performance

julia> @benchmark besselj0(x) setup=(x=130.0 + rand())
BenchmarkTools.Trial: 10000 samples with 999 evaluations.
 Range (min … max):  8.550 ns … 34.869 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     8.675 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   8.688 ns ±  0.572 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

                         ▄          █                         
  ▂▁▁▁▁▁▁▁▁▁▁▄▁▁▁▁▁▁▁▁▁▁▁█▁▁▁▁▁▁▁▁▁▁█▁▁▁▁▁▁▁▁▁▁▁▄▁▁▁▁▁▁▁▁▁▁▃ ▂
  8.55 ns        Histogram: frequency by time        8.76 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

## bad
julia> @benchmark besselj0(x) setup=(x=1e17 + rand())
BenchmarkTools.Trial: 10000 samples with 994 evaluations.
 Range (min … max):  30.600 ns … 59.397 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     30.768 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   30.845 ns ±  0.856 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

    ██                                                         
  ▃▇███▄▂▂▂▂▂▁▁▁▁▂▁▁▁▁▁▁▂▂▁▂▂▁▁▁▂▁▂▂▁▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▂▂▂▂▂ ▂
  30.6 ns         Histogram: frequency by time        33.5 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

Perhaps we could improve this whole function a little bit but I remember the problem being with rem_pio2 in Julia. Could revisit this.

At large arguments besselj0(x) ~ sqrt(2/pi) * cos(x - pi/4) / sqrt(x) so we should be able to do a lot better here then our current approach. We could potentially also improve the range of validity of this approximation by adding a term in the cos term besselj0(x) ~ sqrt(2/pi) * cos(x - pi/4 - 1/8x) / sqrt(x).
The polynomial outside of the cos quickly becomes one but the periodic nature of the cos function requires us to keep some more terms except for very large arguments.

No matter what though, computing cos(x + y) accurately is the important thing here. But we should be able to do much better because this is such an important function.

[heltonmc]

Closed by #88