#109 pdf.c needs downscaling logic

[Alan W. Irwin]

The scaling between internal PLplot "physical" coordinates and actual HPDF coordinates is bad as evidenced by the range of physical coordinates being substantially smaller than the optiimum (PIXELS_X-1, PIXELS_Y). There is also a very large canvas size on the HPDF side things corresponding to 50 inches by 37.5 inches (!) which is probably also a sign of scaling troubles. The solution (used by e.g., modern device drivers like svg.c, qt.cpp, and cairo.c) is to downscale the coordinate transform between the PLplot physical coordinates and the native coordinates of the device. However, it is a little tricky to implement because the actual scaled HPDF coordinates are only known after the final size of the canvas is known which occurs in the bop (beginning of page) routine. So I am assigning this issue to Werner (with his consent) since he is most familiar with this driver's logic. I am leaving it as the same priority as the next pdf.c bug I am about to post since that bug could conceivably be caused by bad scaling.