primal-estimate: use optimal estimates across u64 ranges, add better estimates

[huonw]

This improves the performance (that is, the distance between the lower and upper bound) of nth_prime and prime_pi.

• prime_pi improves up to 22% for numbers between 56 and 3400
• prime_pi improves from 24% to 80% for numbers larger than 185000 (more improvement for large numbers
• nth_prime improves from 28% to 87% for numbers larger than 95000 (again, more improvement for large numbers)

This is done in two ways:

• adding more estimates from http://arxiv.org/abs/1002.0442 (fixes https://github.com/huonw/primal/issues/23)
• creating a generator that computes the estimates at many points in the u64 range, and selects the tightest ones (printing out match arms covering the subranges for which each generator is the best)

[autohuonw]

@huonw: no appropriate reviewer found, use r? to override

[codecov-io]

Codecov Report

Merging #26 into master will decrease coverage by 1.03%. The diff coverage is 23.8%.

@@            Coverage Diff            @@
##           master     #26      +/-   ##
=========================================
- Coverage   52.24%   51.2%   -1.04%     
=========================================
  Files          16      17       +1     
  Lines        3857    3933      +76     
=========================================
- Hits         2015    2014       -1     
- Misses       1842    1919      +77

Impacted Files	Coverage Δ
generators/src/bin/wheel-generator.rs	0% <ø> (ø)
generators/src/bin/estimate-generator.rs	0% <0%> (ø)
primal-estimate/src/lib.rs	96.36% <92.59%> (+0.86%)	:arrow_up:
primal-sieve/src/lib.rs	100% <0%> (ø)	:arrow_up:

---

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 e7fbe98...c98724f. Read the comment docs.

[huonw]

Thanks for catching those nits. I'm unsure about fuzzing tests, as the relationships between the final assertion and the input is rather non-obvious so fuzzers probably won't be as effective as other techniques (e.g. testing exhaustively), but definitely something I'll think about.

bors r=cuviper

[bors[bot]]

Build succeeded

• continuous-integration/travis-ci/push