It is annoying since in relationship to bounds checking, we could write a non-surprising automatic conversion - given array length type and index type, we'd need to cast to the bigger of the two and make the bound check. But it doesn't mesh well with the fact that all but primitive indexing are implemented in the library, and is not built in.
This is indeed the main reason array[42u32] isn't supported today: implementing Index on both usize and u32 would make array[42] ambiguous for the type system, it needs to unify the {integer} literal and some implementor of Index, but there are now two possibilities. This would be a breaking change.
