[WebGPU] Plug memory leaks and free resources on shutdown

[nikhilJain17]

This diff frees wgpu::Buffers and in buffer pools on shutdown to prevent memory leaks on GPU. It also fixes memory leaks on the heap, where we allocate backend, backend_ctx, buffer_ctx, and decisions on the heap but never delete them. These are either explicitly deleted (wrt ggml lifecycle) or changed to be smart pointers.

We implement destructors for our buffer pool structs, webgpu_context struct and webgpu_global_context struct. Since webgpu_global_context is a refcounted smart pointer, it will destruct automatically when all thread contexts have been destroyed.

We call free on all the buffers we allocate, and we explicitly free our buffer pools and debug/error/staging buffers.
Also, since we explicitly wait on all our callbacks, we do not have to worry about waiting for callbacks while shutting down.

Memory leak on the heap before.

No memory leak on the heap after.

[nikhilJain17]

I changed this into a shared_ptr because this was leaking on the heap since we never deallocated it.

[nikhilJain17]

I believe other wgpu members, like Instance, Device, Queue, and Pipeline, are refcounted and delete automatically when all references are deleted. But Buffers need to be explicitly destroyed.

[nikhilJain17]

Also since webgpu_global_context is a shared_ptr, its destructor is automatically called once all references to it are deleted.

[nikhilJain17]

These are both allocated on the heap and leak if we don't delete them. Maybe we could turn them into shared_ptr but I don't know how it would behave once the pointer is passed around in the ggml lifecycle.