Some users have requested greater flexibility in setting scores beyond the current range of -1,000,000 to +1,000,000. Usually this is to be able to use some external value directly.
We could raise INFINITY to a larger number (possibly expanding the type to 64-bit integer), and/or allow floating-point scores. Alternatively, we could define a score structure containing a numeric value and flags for +/- INFINITY.
I lean against this one, because there's no guarantee an external floating-point value can be exactly represented in a C type, and because INFINITY should ideally would be less than half the range of the type used to store it, to allow bare arithmetic addition of two values (meaning we couldn't allow full use of any integer type's range).
This would break backward compatibility in unpredictable ways, so it should be done at a major series bump.