If we are comparing 2 numbers that are even integers, compare them via
big.Int values rather than formatted strings. This avoids problems with
large integers stored in big.Float with different precisions which could
format differently.

There is a bug in the compact floats implementation, which could cause
some integer values stored in a float to encode as an incorrect value.
The result of the expression used to check for compact integer encoding
is `float64(int64(n)) == n`, which is undefined when conversion is
not exact.