GPU support by adding CuParticles

[baggepinnen]

This PR adds a type CuParticles that works as regular Particles but is backed by a CuArray. Expect in the range of 5-7x speedup over Particles if the sample size is 1e5 - 1e7.

EDIT 2021-04-22: For complex functions, the benefit of CuParticles may appear already above 1e3 particles.

Future memory-management improvements to the GPU ecosystem is likely to improve upon this number.

[codecov[bot]]

Codecov Report

Merging #80 (f42a6e7) into master (fb61791) will decrease coverage by 0.53%. The diff coverage is 88.32%.

@@            Coverage Diff             @@
##           master      #80      +/-   ##
==========================================
- Coverage   94.27%   93.73%   -0.54%     
==========================================
  Files          18       19       +1     
  Lines         891      990      +99     
==========================================
+ Hits          840      928      +88     
- Misses         51       62      +11     

Flag	Coverage Δ
unittests	93.73% <88.32%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/MonteCarloMeasurements.jl	88.23% <0.00%> (-2.68%)	:arrow_down:
src/bymap.jl	98.24% <ø> (ø)
src/types.jl	90.90% <0.00%> (-4.33%)	:arrow_down:
src/particles.jl	96.32% <85.00%> (-1.23%)	:arrow_down:
src/plotting.jl	90.57% <87.01%> (+1.04%)	:arrow_up:
src/complex.jl	84.37% <100.00%> (ø)
src/deconstruct.jl	91.97% <100.00%> (+0.25%)	:arrow_up:
src/nominal.jl	100.00% <100.00%> (ø)
src/register_primitive.jl	100.00% <100.00%> (ø)
src/resampling.jl	100.00% <100.00%> (ø)
... and 3 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update fb61791...f42a6e7. Read the comment docs.