gcode test 3 fails on 32-bit architectures

Girvin reported after the release of 4.1.3 that gcode test #3 was failing. This test uses a higher "dpi" to generate the gcode output. This needs to be investigated.

I haven't really looked very hard yet, but, it seems that there is a floating point precision issue here. I'm guessing it is in a trig function somewhere. The precision of the output isn't particularly high, so, I'm surprised to see an issue.

Girvin is running this on a 32-bit AMD Athalon II.

From his message:
==================================== Snip ====================================
Running Slackware Linux 14.2 (K4.4.157) & xfce4.

FYI: I am still getting one test failure - hid_gcode3 - when I do a make
check. Here is a snippet from the make output:

Test: hid_gcode3
(cd outputs/hid_gcode3 && ../../../src/pcbtest.sh -x gcode --dpi 1200
gcode_oneline.pcb)
--- /tmp/pcb_tests.9260/gcode_oneline-bottom.gcode-ref 2018-10-04
14:14:15.000000000 -0700
+++ /tmp/pcb_tests.9260/gcode_oneline-bottom.gcode-out 2018-10-04
14:14:15.000000000 -0700
@@ -36,6 +36,6 @@
  G1 X28.765500 Y12.488333
  G1 X28.765500 Y12.890500
-G1 X28.638500 Y13.165667
-G1 X28.448000 Y13.377333
+G1 X28.617333 Y13.208000
+G1 X28.405667 Y13.398500
  G1 X28.130500 Y13.525500
  G1 X27.813000 Y13.546667
--- /tmp/pcb_tests.9260/gcode_oneline-top.gcode-ref 2018-10-04
14:14:15.000000000 -0700
+++ /tmp/pcb_tests.9260/gcode_oneline-top.gcode-out 2018-10-04
14:14:15.000000000 -0700
@@ -36,6 +36,6 @@
  G1 X23.706667 Y12.488333
  G1 X23.706667 Y12.890500
-G1 X23.579667 Y13.165667
-G1 X23.389167 Y13.377333
+G1 X23.558500 Y13.208000
+G1 X23.346833 Y13.398500
  G1 X23.071667 Y13.525500
  G1 X22.754167 Y13.546667
FAIL
==================================== Snip ====================================