Improve performance of ALMA()

Ethan Smith pointed out significant performance improvement from
replacing the rollapply() call with WMA().

   Unit: milliseconds
          expr       min         lq      mean     median        uq     max neval
       ALMA(z) 10.692501 11.2338510 12.426196 11.5029515 11.944251 33.3690   100
   ALMA.new(z)  0.403302  0.4692515  0.688744  0.5237505  0.575451 17.0059   100

The rollapply() solution handled multi-column inputs. So initialize
the result and loop over the columns. Add a try.xts() at the beginning
of the function to handle cases where 'x' is a vector.

Closes #117.

R/MovingAverages.R

@@ -432,6 +432,7 @@ function(x, n=20, ...) {
 function(x, n=9, offset=0.85, sigma=6, ...) {
 
   # ALMA (Arnaud Legoux Moving Average)
+  x <- try.xts(x, error=as.matrix)
 
   if(offset < 0 || offset > 1) {
     stop("Please ensure 0 <= offset <= 1")
@@ -445,7 +446,10 @@ function(x, n=9, offset=0.85, sigma=6, ...) {
   sumWeights <- sum(wts)
   if(sumWeights != 0)
     wts <- wts/sumWeights
-  alma <- rollapply(x, width=n, FUN=function(xx) sum(xx*wts),
-                    fill=NA, align="right")
+
+  alma <- x * NA_real_
+  for(i in seq_len(NCOL(x))) {
+    alma[,i] <- WMA(x[,i], n, wts)
+  }
   reclass(alma, x)
 }