Non-Linear Regression with Brute Force

Adds brute force and multiple starting values to nls.

The nls2 package has a single function, nls2, which can be used to 
sample an nls formula using grid, uniform random or Latin Hypercube sampling
returning the best result or all of them.  This can be useful for
generating starting values for nls.

It can also evaluate an nls formula at a point where the Jacobian is singular
(which cannot be done in nls).  This can be used to identify which parameters
are troublesome.  (See example 4 in the nls2 help file.)

It is easy to use for those familiar with nls since it has the same set of
arguments plus all=.  The algorithm= and start= arguments are expanded to
allow specificaiton of the additional functionality.

See the INSTALL file for installation instructions and the ?nls2 help file
for detailed documentation.