Eliminate enclose argument

[hadley]

This eliminates one argument we need to thread through multiple function calls by creating a new environment with appropriate parent. (I doubt anyone actually uses this argument but it should be a safe transformation)

[cderv]

Seems ok.

We need to update knitr sew.warning function so that the enclos is not expected anymore.

Currently it is

#' @export
sew.warning = function(x, options, ...) {
  call = if (is.null(x$call)) '' else {
    call = deparse(x$call)[1]
    if (call == 'eval(expr, envir, enclos)') '' else paste(' in', call)
  }
  msg_wrap(sprintf('Warning%s: %s', call, conditionMessage(x)), 'warning', options)
}

Source: https://github.com/yihui/knitr/blob/e1edd34f4149d996efce5cf539e781eda6e3b86a/R/output.R#L557

For context on this, I am running your PR against knitr-examples repo and knitr test suits to see which impact it could have. It helps me find the possible problems like this one

Diff we get from knitr-examples

--- a/007-text-output.md
+++ b/007-text-output.md
@@ -477,7 +477,7 @@ warning("no no no")
 ```
 
 ```
-## Warning: no no no
+## Warning in eval(expr, envir): no no no
 ```
 
 Invalid indices will select nothing:

[hadley]

Hmmmmm, I think we should be able to fix that in evaluate() so knitr continues to work, and you can eventually remove the existing code. I'll file a new issue momentarily.