fix #H261 - bias in Bresenham's algorithm in walk_path

From a bug report, walk_path() was
favoring orthogonal steps at beginning of the path and diagonal ones at
end, when intermixing diagonal and orthoganal produced a more accurate
representation of the real path.  Only mattered for long distance jumps
3x2 or 3x1 steps away; hurtling always moves in a straight line and short
(2x1 knight) jumps aren't affected by the patch supplied with the report.

doc/fixes34.4

@@ -342,6 +342,7 @@ lit candle or potion of oil which burned out while equipped would leave stale
 wielded/worn figurine which auto-transformed had same stale pointer bug
 format names of not yet id'd artifacts such that obj type shows for non-weapons
 make quest leader and nemesis be unlikely to be affected by traps
+use a more precise jumping path for far, non-straight line destinations
 
 
 Platform- and/or Interface-Specific Fixes

src/dothrow.c

@@ -433,10 +433,10 @@ walk_path(src_cc, dest_cc, check_proc, arg)
 	    prev_x = x;
 	    prev_y = y;
 	    y += y_change;
-	    err += dx;
-	    if (err >= dy) {
+	    err += dx << 1;
+	    if (err > dy) {
 		x += x_change;
-		err -= dy;
+		err -= dy << 1;
 	    }
 	/* check for early exit condition */
 	if (!(keep_going = (*check_proc)(arg, x, y)))
@@ -447,10 +447,10 @@ walk_path(src_cc, dest_cc, check_proc, arg)
 	    prev_x = x;
 	    prev_y = y;
 	    x += x_change;
-	    err += dy;
-	    if (err >= dx) {
+	    err += dy << 1;
+	    if (err > dx) {
 		y += y_change;
-		err -= dx;
+		err -= dx << 1;
 	    }
 	/* check for early exit condition */
 	if (!(keep_going = (*check_proc)(arg, x, y)))