Optimize `Vec::split_off`

Now that we have our own Vec implementation, we can add some APIs which were previously impossible.

Vec::split_at can convert a Vec into 2 x Vecs, without copying any data or making any allocations.

let original_len = vec.len();
let original_capacity = vec.capacity();
let original_ptr = vec.as_ptr();

// `split_at` returns `None` if `vec.len() < 2`
let (vec1, vec2) = vec.split_at(2).unwrap();

assert!(vec1.len() == 2);
assert!(vec1.capacity() == 2);
assert!(vec2.len() == original_len - 2);
assert!(vec2.capacity() == original_capacity - 2);
// `vec1` and `vec2` point to chunks of the original `vec`
assert!(vec1.as_ptr() == original_ptr);
assert!(vec2.as_ptr() == original_ptr.add(2));

This would be useful in various places in transformer where we currently have to allocate new Vecs and copy data around. e.g. https://github.com/oxc-project/oxc/pull/10434#issuecomment-2809350194

Start by locating the repository's own Vec implementation and the existing split_at entry point. Use the stated behavior as the acceptance criteria: two Vecs should share the original allocation without copying, with the specified lengths, capacities, and pointers; add or run focused tests for those properties.