Take some lowhanging fruit to speed up R6RS fixnum operations

[Andreas Rottmann]

In porting dorodango[0], I have noticed that Guile's R6RS fixnum
operations are quite slow; here's a patch to remedy that a bit.  The
patch contains a benchmark for `fxxor', which experiences a performance
improvement of >2 with the patch.

[0] http://home.gna.org/dorodango/

I also used real code to validate the performance impact, namely Göran
Weinholt's ZIP implementation from his industria collection[1]. There
the patch makes for an ~1.7x improvement for uncompressing a 675KiB
(compressed) ZIP file containing text files that yield good compression.

[1] https://code.launchpad.net/~weinholt/scheme-libraries/industria

However, it must be noted that while this patch improves things,
uncompressing the aforementioned ZIP file takes still 42 seconds on my
machine (down from 74 before).  I have another patch in the queue to
improve this by making `fixnum?' a VM primitive, but even with that,
performance is not stellar.