To save a bit of consing and calls to atan2,

        * numbers.c (scm_angle): Use scm_flo0 for non-negative inum, bignum
        and real.

I guess the same can be done for negatives, returning pi.  Would
scm_sys_protects be the right place to add an scm_flo_pi?  I see
abs_most_negative_fixnum uses scm_permanent_object instead.

For the actual pi value, I guess there'd be a choice between M_PI and
the return from atan2(0.0,-1.0).  The two ought to be the same of
course.  Maybe use the constant and have one of the tests see that
atan2 agrees.