Improve speed of Tk_PhotoPutBlock for one-row case

In my application, I use tkimg1.3 to load xpm images
with transparency information.

I compared the time taken to load the image in
tk8.2+tkimg1.3 to the time taken in tk8.4+tkimg1.3:
8.2: 59086 microseconds
8.4: 769918
8.4, patched: 77059

This patch expands the 'single pixel' case into a
'single span' case: one or more pixels on a single row
of the image, where all have alpha 0 or 255. (in fact,
tkimg1.3's xpm loader always sends spans with fully
opaque pixels)

As you can see, this patch reduces the time from 13x as
long as 8.2 to only 1.3x as long as 8.2.