Install Nemo and cache it

[fingolfin]

No description provided.

[codecov]

Codecov Report

Merging #99 into master will decrease coverage by 0.71%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master      #99      +/-   ##
==========================================
- Coverage   51.38%   50.66%   -0.72%     
==========================================
  Files          33       37       +4     
  Lines        1919     2481     +562     
==========================================
+ Hits          986     1257     +271     
- Misses        933     1224     +291

Impacted Files	Coverage Δ
JuliaExperimental/gap/gapnemo.g	97.91% <0%> (ø)
JuliaExperimental/gap/realcyc.g	85.18% <0%> (ø)
JuliaExperimental/gap/numfield.g	30.99% <0%> (ø)
JuliaExperimental/gap/hnf.g	42.85% <0%> (ø)
JuliaExperimental/gap/arith.gi	77.04% <0%> (+1.63%)	⬆️
JuliaExperimental/read.g	96.77% <0%> (+15.29%)	⬆️
JuliaExperimental/julia/realcyc.jl	86.66% <0%> (+86.66%)	⬆️
JuliaExperimental/julia/hnf.jl	100% <0%> (+100%)	⬆️

[fingolfin]

To my surprise, installing Nemo.jl only adds ~10 minutes to the Travis build.

[fingolfin]

This seems to be working fine now. There is the question of what happens if the Julia code in Nemo "should be" recompiled due to a change in the Julia nightly... not sure whether Julia does that automatically (probably not?), or else how to trigger it manually... I think (but might be wrong) that the part taking up most of the time is compiling all the C libraries, so as long as we can avoid that, it should still be useful. Perhaps @wbhart or some other Julia experts have some insights on that?

[wbhart]

I believe Julia will recompile if its binary changes. However, it won't recompile Flint and so on, I think, as this shouldn't be needed. You are right that the first build takes a while, due to building all the C libraries. We surely want to eventually move over to using binaries to speed that up.

[sebasguts]

Assumning Bills comment, this is fine for now.

You can merge it if you want to.