Use %uvector-replace in replace even when subseqs differ in length

[se-mz]

This significantly speeds up code that relies on the way replace implicitly intersects sequence bounds. For example, the replace calls below take about 150ms & 3ms respectively before the patch; both take about 3ms afterwards.

(let ((buf  (make-array 100000000 :element-type '(unsigned-byte 8)))
      (buf2 (make-array 10000000  :element-type '(unsigned-byte 8) :initial-element 1)))
  (time (replace buf buf2))
  (time (replace buf buf2 :start1 0 :end1 (length buf2)))
  nil)

Originally found by Gilbert Baumann a while ago.

[xrme]

In the case of something like
(make-array 100000000 :element-type '(unsigned-byte 8)), where the memory is implicitly zero-filled (in CCL, anyway), it is possible that the memory will not actually be "real" yet, and that therefore the make-array call will take very little time. (The memory will be provided zero-filled on demand pages from the operating system when we're consing into fresh memory.)

In a fresh lisp. you'll almost certainly see a time difference between (make-array 100000000 :element-type '(unsigned-byte 8)) and (make-array 100000000 :element-type '(unsigned-byte 8) :initial-element 1).

Your improvement to replace is probably a win anyway, but I wanted you to be aware of a possible source of timing error because you might accidentally be measuring the time needed for the operating system to provide zero-filled-on-demand pages for buf in the first replace form.

[se-mz]

Good catch, but it seems like that doesn't matter here; I get comparable readings when initializing both with non-zero values.