using expressions as the input vector in parallel_reduce

[paciorek]

Using an expression for the input vector currently fails. I didn't dive too far trying to figure out where to intervene to, presumably, lift the expression into a temporary, as it seemed most efficient to get some guidance from @perrydv .

We might decide to punt on this (though I expect it won't be hard to address), in which case we should/can easily error trap this in an informative fashion.

   nc <- nClass(
    Cpublic = list(
      go = nFunction(
        fun = function(x = 'numericVector',y='numericVector') {
          z <- parallel_reduce('pairmin', x+y, 100)
          return(z)
        },
        returnType = 'numericScalar'
      )
    )
  )

>  Cnc <- nCompile(nc)
To-do: Care needed to filter interfaced methods by exportMembers names.
Error: No variable named: +

I'll also note as related that with some machinations one can break parallel_for with expressions for the extent of the looping, such as the following, which fails at C++ compilation. (However, using simply length(x) is fine, so this is an edge case, I think.)

nc <- nClass(
    Cpublic = list(
      go = nFunction(
        fun = function(x = 'numericVector') {
            y <- x
           two <- 2
          parallel_for(i, 1:length(x+y),  
                       {y[i] <- 2 * x[i]},
                       c("x"), ## copy for each thread
                       c("y")) ## share across threads
          return(y)
        },
        returnType = 'numericVector'
      )
    )
  )