FMSLogo Code

Fix a copy&paste error that prevented the second line of CSaveBeforeExitDialog from being wrapped. This didn't matter in practice because the second line is shorter than the wrap-size (at least in English)

Add some comments to CWorkspaceEditor::OnClose() which explain why code calls endedit() twice.

Fix a regression that was introduced by moving the delayed Close() call from the idle event to stop_execution() that made FMSLogo think that the workspace was dirty if you halted a program during its "startup" execution. Since it's common for a program to never return from its "startup", this regression made it so that FMSLogo would give a strange warning when any program was running.

The fix was to restore the "IsDirty" flag in-between when a file is loaded and when its "startup" list is run, instead of after the startup list is run.

Implement feature #33; Make it so that FMSLogo doesn't mind exiting while running and remove the "Logo doesn't like to exit while running" dialogs.

This involved moving the delayed Close() invocation from the idle loop to the point where the evaluation engine has fully unwound. Then some logic was added to the OnClose() to halt the language engine if it is running.

Integrate a regression test for bug #484 into the full test pass.

Simplify the exit path by removing prepare_to_exit(), which was legacy from UCBLogo that didn't make sense in FMSLogo and only obscured the shutdown mechanics.

Add a regression test that actually runs BYE (by running it in a child instance)

Include logodialogboxes.cpp in the GTK build.
This was primarily done to get support YESNOBOX to unblock the tests,
but they quickly blocked on GIFSAVE, which was immediately after the
YESNOBOX call.

Fix a bug that caused FORWARD to stop drawing lines after POLYVIEW was run
on GTK2 versions of wxWidgets (that is, on GNU/Linux). The bug was a
reference counting bug that caused POLYVIEW to destroy the turtle's pen.

Fix a bug that caused FORWARD to stop drawing lines after BITBLOCK
was run on GTK2 (GNU/Linux). In retrospect, the bug is obvious,
by maintaining a const reference to the wxPen that lbitblock() saves
and restores, the internal reference count is not incremented and
the wxPen is destroyed when it's replaced with a transparent pen,
leaving the saved reference to refer to an invalid pen.

Putting the saved pen into a new wxPen object fixes the problem.

I don't know why this problem doesn't affect MSW builds.