Too restrictive return type in std.math.min

[Ekdohibs]

Since 2cd5e55, the return type of std.math.min is made more restrictive in some situations. However, it seems to me that it is made too restrictive: if A and B are both signed ints (for instance, i16 and i32), then the most restrictive type able to contain the result is the widest of the two. Indeed, in the above case, the minimum might be as little as -2^31, which does not fit in an i16.
Likewise, floats are always signed and the return type should be the widest of the two.
I am not sure about comptime ints and floats: should the return type depend on the comptime value?

[andrewrk]

Thanks for the report. This is a good argument in favor of inferred return types (#447).