stdlib: Use builtin tail-allocated arrays instead of ManagedBuffer to allocate contiguous array buffers.
This reduces the amount of SIL generated for array operations significantly.
The generated code should be mostly the same (modulo different inlining decisions).