We currently use a Linear Allocator. It's very inefficient and means we need ~128 x 64KB pages to validate a block with a 350kb transaction in it. Switch this for a Buddy allocator or Slab allocator.

Furthermore, sr-io::storage was recently pessimised through adding a needless memcpy in to_vec. The code should be reverted thus:

/// Get `key` from storage and return a `Vec`, empty if there's a problem.
pub fn storage(key: &[u8]) -> Option<Vec<u8>> {
	let mut length: u32 = 0;
	unsafe {
		let ptr = ext_get_allocated_storage(key.as_ptr(), key.len() as u32, &mut length);
		if length == u32::max_value() {
			None
		} else {
			Some(<Vec<u8>>::from_raw_parts(ptr, length as usize))
		}
	}
}

To ensure this code doesn't result in UB on the part of Vec, ext_get_allocated_storage should ensure that the pointer is aligned according to u8 allocations on the unknown-unknown-webassembly platform (almost certainly 4 byte). This is won't be necessary as long as the Heap implementation ensures all allocations are on such a boundary.