new challenges in scons cleaning, things basically ok

[Greg Troxel]

I have been trying to test and getting odd failures that don't seem to
quite reproduce.   I tracked that down to a new cache file, not removed
by scons --clean, appearing:
  .sconsign.4.dblite
instead of the previous .sconsign.dblite.  I added a workaround to  my
build script to remove this after invoking scons --clean.


I think I have found at least one and may two minor bugs, which I
suspect is longstanding and probably best not to touch for this release.
The clearest one is that when passing libgpsmm=false, the build tries to
creat a qgpsmm test program, which can't link bcause libgpsmm doesn't
exist.

So qt should probably be forced off if libgpsmm is off. 

The other maybe-problem is that the provided LDFLAGS seem not to get
passed to the build of this test program, resulting in it not finding
some gtkmm stuff.   I realize that's fuzzy and will look into it post
release.


I'm now building with libgpsmm=false qt=false and tests are running ok
so far.