Allow passing epsilon_factor in gradient functions

[andreasnoack]

It can be useful when the objective function is computed in lower precision e.g. when it's a numerical solution to an ODE.

[coveralls]

Coverage increased (+15.3%) to 57.143% when pulling 4d5fac9 on andreasnoack:an/eps into 21eba38 on JuliaDiffEq:master.

[coveralls]

Coverage decreased (-1.6%) to 40.201% when pulling 2416927 on andreasnoack:an/eps into 21eba38 on JuliaDiffEq:master.

[coveralls]

Coverage increased (+15.3%) to 57.143% when pulling 4d5fac9 on andreasnoack:an/eps into 21eba38 on JuliaDiffEq:master.

[ChrisRackauckas]

I seems like this should be a uniform change, i.e. should be done to the derivative and Jacobian functions as well?

[codecov]

Codecov Report

Merging #52 into master will decrease coverage by 0.45%.
The diff coverage is 90%.

@@            Coverage Diff             @@
##           master      #52      +/-   ##
==========================================
- Coverage   58.47%   58.02%   -0.46%     
==========================================
  Files           6        6              
  Lines         301      293       -8     
==========================================
- Hits          176      170       -6     
+ Misses        125      123       -2

Impacted Files	Coverage Δ
src/derivatives.jl	70.83% <100%> (+0.83%)	⬆️
src/jacobians.jl	82.75% <100%> (-0.58%)	⬇️
src/gradients.jl	51.53% <80%> (-0.57%)	⬇️

---

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 21eba38...2416927. Read the comment docs.

[andreasnoack]

Makes sense. I've added the keyword argument to the remaining methods.