Wrong results from in_range() tests with infinite offset

Dean Rasheed pointed out that in_range for float4/float8 seems to be
doing the wrong thing for infinite offsets, and after some testing
I concur that it is. For example, a sort key of '-infinity' should
be considered to be in-range for a range specified as RANGE BETWEEN
'inf' PRECEDING AND 'inf' PRECEDING; but with the code as it stands,
it isn't. I propose the attached patch, which probably should be
back-patched.

When the current row's value is +infinity, actual computation of
base - offset would yield NaN, making it a bit unclear whether
we should consider -infinity to be in-range. It seems to me that
we should, as that gives more natural-looking results in the test
cases, so that's how the patch does it.

regards, tom lane