Re: sconsign database discussion

oh, one more little bit on this sconsign topic...

The current model of dumping the whole db at the end, replacing the 
previous on-disk copy, has what to me seems an odd side effect:

if file modes (permissions) indicate .sconsign.dblite is not writable, 
it will still be overwritten (replaced).  To me, that's a little odd, 
but evidently intentional; a change I made in PR 3837 tripped over this 
because we have a contrived testcase that sets the permissions on the 
file to 0o444, and then expects the file will still be replaced.  I'm 
not going to just change what's clearly long-standing behavior, just 
curious about this expectation.

This is not the same as my earlier question of a database that's been 
explicitly opened read-only.