Remove gradient storage

[anriseth]

Fixes #47

I also updated the README and the tests (can't enable them until Optim 1.0 is tagged)

[codecov]

Codecov Report

Merging #48 into master will decrease coverage by 3.1%.
The diff coverage is 24.59%.

@@            Coverage Diff             @@
##           master      #48      +/-   ##
==========================================
- Coverage   48.63%   45.52%   -3.11%     
==========================================
  Files           8        8              
  Lines         584      615      +31     
==========================================
- Hits          284      280       -4     
- Misses        300      335      +35

Impacted Files	Coverage Δ
src/api.jl	0% <ø> (ø)	⬆️
src/basic.jl	100% <100%> (ø)	⬆️
src/backtracking.jl	75% <100%> (ø)	⬆️
src/morethuente.jl	53.71% <100%> (-0.27%)	⬇️
src/deprecate.jl	27.08% <2.77%> (-72.92%)	⬇️
src/hagerzhang.jl	32.53% <33.33%> (-0.48%)	⬇️
src/strongwolfe.jl	75.67% <80%> (+0.35%)	⬆️

---

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 5eebe50...7436f53. Read the comment docs.

[pkofod]

If you keep a signature with a warning and g_scratch, and make sure that copy!(g_scratch, d.g at the end of this method, like

function _hagerzhang!(df, x, s, xtmp, g, lsr, c, mayterminate, ...)
    warn("you no longer have to provide a "g" input")
    _hagerzhang!(df, x, s, xtmp, lsr, c, mayterminate, ...)
    copy!(g, d.g)
end

then this doesn't have to be a breaking change... don't know if it's worth the trouble though.

[anriseth]

Good idea! I'll try to do this.

[anriseth]

This can be merged, if nobody objects.

Then we can tag a new minor

[pkofod]

I'll merge after the tests run (can't be sure the conflict resolution worked flawlessly)