Optimize writes to `Array` such that they don't read and drop existing values if this isn't necessary

[yorickpeterse]

Description

When writing to an Array using methods such as Array.set, we read the existing value and drop it before the write. For types that are copy (e.g. Int) this is redundant and a waste of CPU time.

We should optimize this such that for copy types we don't perform a read, instead just overwriting the data as-is. Ideally this is implemented in a generic way that works for other types as well. This probably means we need to look for reads of Pointer[T] values where T: copy, then remove such reads if their result isn't passed around. This in turn may depend on some form of lifetime/escape analysis that we'd likely also need for #776

Related work

No response

[yorickpeterse]

Note that while LLVM may also do this for us (I'm not sure), I think it's still worth doing ourselves as it means we depend less on LLVM for optimizations, which in turn might make it easier to switch to a different backend at some point in the future (or add additional backends).