update.default bugfix (PR#3288)

According to the man page for formula, "a formula object has an associated
environment".  However, update.default doesn't use this environment, which
creates problems like the following:

  make.model <- function(x) { lm(medv~.,x) }
  library(MASS)
  data(Boston)
  fit = make.model(Boston)
  fit = update(fit,".~.-crim")
  # Object "x" not found

Update isn't going to satisfy everyone because people want it to do
incompatible things.  However, I don't think this is one of the things it
should be able to do.

In general the environment of a formula is used to look up variables in
the formula, not the data= argument.  (It's also used to look up the
weights= and subset= arguments, but that is a horrible wart that we would
like to excise).

	-thomas