Add ufunc and function dispatchers

[mhostetter]

This PR creates ufunc and function dispatchers. These are classes that wrap a Python implementation of the ufunc or function. Depending on the field being used, either a JIT-compiled or pure-Python ufunc/function is returned. All the necessary input/output conversion and verification is done.

This cleans up the class clutter. It will also make adding GPU support easier and more convenient.

Additionally, JIT functions now use the lookup ufuncs (if configured). This has a massive speed up for small fields that use lookup tables. Previously, only explicit calculation was used in JIT functions (eg, polynomial multiplication, matrix multiplication, etc). The difference is most notable in GF(p^m) fields.

Before

In [1]: import galois

In [2]: GF = galois.GF(7**5)

In [3]: f = galois.Poly.Random(100, seed=1, field=GF)

In [4]: g = galois.Poly.Random(10, seed=2, field=GF)

In [5]: f * g
Out[5]: Poly(3419x^110 + 11806x^109 + 16253x^108 + 11624x^107 + 7122x^106 + 10439x^105 + 9726x^104 + 16156x^103 + 14118x^102 + 11739x^101 + 5479x^100 + 9674x^99 
+ 14884x^98 + 13054x^97 + 13987x^96 + 14889x^95 + 528x^94 + 4033x^93 + 5662x^92 + 5326x^91 + 16504x^90 + 4999x^89 + 3070x^88 + 2773x^87 + 14723x^86 + 15849x^85 + 13785x^84 + 11438x^83 + 5109x^82 + 463x^81 + 1497x^80 + 6584x^79 + 11814x^78 + 7924x^77 + 7464x^76 + 7160x^75 + 11391x^74 + 12445x^73 + 5249x^72 + 742x^71 + 9192x^70 + 14644x^69 + 7194x^68 + 3017x^67 + 16424x^66 + 1683x^65 + 11817x^64 + 10766x^63 + 10693x^62 + 13717x^61 + 4815x^60 + 10120x^59 + 15956x^58 + 16740x^57 + 3723x^56 + 9490x^55 + 13519x^54 + 9782x^53 + 309x^52 + 11633x^51 + 8786x^50 + 3733x^49 + 7210x^48 + 10879x^47 + 7836x^46 + 4835x^45 + 15762x^44 + 6813x^43 + 15021x^42 + 5566x^41 + 6210x^40 + 4221x^39 + 11845x^38 + 6928x^37 + 5771x^36 + 6221x^35 + 2355x^34 + 1316x^33 + 11194x^32 + 14704x^31 + 3411x^30 + 272x^29 + 16736x^28 + 15618x^27 + 6033x^26 + 13613x^25 + 11378x^24 + 61x^23 + 11063x^22 + 603x^21 + 2506x^20 + 13488x^19 + 702x^18 + 1595x^17 + 6065x^16 + 5784x^15 + 202x^14 + 16275x^13 + 4924x^12 + 1080x^11 + 462x^10 + 11150x^9 + 1046x^8 + 13376x^7 + 15037x^6 + 10448x^5 + 12167x^4 + 5285x^3 + 2705x^2 + 3725x + 6317, GF(7^5))

In [6]: %timeit f * g
1.34 ms ± 3.17 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [7]: pow(f, 123456789, g)
Out[7]: Poly(4190x^9 + 806x^8 + 5109x^7 + 10008x^6 + 8421x^5 + 3671x^4 + 16697x^3 + 4503x^2 + 1386x + 12192, GF(7^5))

In [8]: %timeit pow(f, 123456789, g)
35.1 ms ± 102 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

In [9]: A = GF.Random((100, 100), seed=1)

In [10]: B = GF.Random((100, 100), seed=2)

In [11]: A @ B
Out[11]: 
GF([[13407,  9380, 11364, ...,  9141,  1002,  4655],
    [ 8455,  1277, 13016, ...,  3880, 15000, 12969],
    [ 3576,  9945,  2328, ...,  6984,  4358,  2169],
    ...,
    [12030,  3755,  8130, ...,  9588,  3756, 14957],
    [ 9023,  2430, 10462, ...,  5847,    96, 14494],
    [11625, 16576,  5221, ...,  8900, 12055,  4516]], order=7^5)

In [12]: %timeit A @ B
1.1 s ± 13.4 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

After

In [6]: %timeit f * g
68 µs ± 131 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

In [8]: %timeit pow(f, 123456789, g)
378 µs ± 1.09 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [12]: %timeit A @ B
17.3 ms ± 56.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

[codecov]

Codecov Report

Merging #354 (1e19b27) into master (bcc8cba) will increase coverage by 0.59%.
The diff coverage is 91.34%.

@@            Coverage Diff             @@
##           master     #354      +/-   ##
==========================================
+ Coverage   92.93%   93.52%   +0.59%     
==========================================
  Files          42       41       -1     
  Lines        5404     5265     -139     
==========================================
- Hits         5022     4924      -98     
+ Misses        382      341      -41     

Impacted Files	Coverage Δ
galois/_polys/_primitive.py	99.19% <ø> (ø)
galois/_fields/_gf2.py	68.00% <64.44%> (-32.00%)	⬇️
galois/_domains/_meta.py	81.48% <81.48%> (ø)
galois/_domains/_array.py	95.42% <82.60%> (+4.86%)	⬆️
galois/_polys/_poly.py	90.82% <84.00%> (ø)
galois/_polys/_dense.py	87.31% <86.80%> (-10.47%)	⬇️
galois/_domains/_ufunc.py	89.47% <89.09%> (-2.15%)	⬇️
galois/_domains/_function.py	83.88% <89.31%> (+0.07%)	⬆️
galois/_lfsr.py	89.60% <89.41%> (+5.23%)	⬆️
galois/_codes/_bch.py	97.60% <95.45%> (-0.46%)	⬇️
... and 13 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 bcc8cba...1e19b27. Read the comment docs.