Updating gradient and Jacobian functions

[aadler]

This will address issue #168. The commit log is a bit wonky since in my experiments I set the five-point estimation as the sole version and then @astamm suggested we make 3/5 a user option. We can do that easily in all cases except when check_derivatives = TRUE is passed in nloptr since all options are overwritten by nloptr.add.default.options/nloptr.get.default.options inside of nloptr and I didn't want to start playing with those. So in the "automatic" case, the five point option is used. Otherwise, the default is three unless the user passes five. Synopsis of changes:

• nl.grad, nl.jacobian, and check.derivatives now all use the same algorithm.
• finite.differences has been removed.
• nloptr passes points = 5 to check.derivatives if the check_derivatives option is set to TRUE.
• heps defaults to NULL and is set to the optimum value (based on Sauer) given points. If a values is passed to heps, that is used.
• Update documentation and unit tests as needed.
• Bump minor version since this is new functionality that does not break the old.
• Unrelated Removed nloptr.show.inequality.warning setting option in zzz.R given release 2.1.0. It should have been removed there; my error.

[aadler]

Team, any thoughts?

[aadler]

@astamm Any objections?