Typing is technically a distinct concept from binary representation, it just happens to be a very useful concept for interpreting bits. If the set of valid operations is known at compile time with no exceptions, I'd call that strong typing regardless of whether the internal representation changes during execution.

The problem I do see with combining int and float is that it means that you always have to be careful with == on any pair of numbers, where a language that does distinguish between the two gives you a cue about when equality will work and when it won't.

But that's a flaw in the number type, not a weakness in the type system.