Reduce size_of Node for better cache performance in render loop

[orottier]

std::mem::size_of::<Node>() == 160 currently. I think we gain some nice performance (better cache utilization) if we trim it down a bit.

Some suggestions:

• Use u8 instead of usize for input/output indices
• Use u32 for NodeIndex instead of u64
• Store outgoing_edges on the heap
• Pack ChannelConfig as a single Arc<(count, mode, interp)> ChannelConfig is now a single Arc instead of 3 #307
• Use bitflags to combine the boolean fields in a single u8

On a different note, we can explore to store the Nodes as an "struct of arrays" instead of "array (hashmap in our case) of structs" https://news.ycombinator.com/item?id=17981306

[b-ma]

On a different note, we can explore to store the Nodes as an "struct of arrays" instead of "array (hashmap in our case) of structs" https://news.ycombinator.com/item?id=17981306

Very nice direction to dig I think, kind of related to this comment there #55 (comment) (in a more interesting way :)

[orottier]

For the record, in the current implementation:

        assert_eq!(160, std::mem::size_of::<RefCell<Node>>());

        assert_eq!(8, std::mem::size_of::<RefCell<()>>());

        assert_eq!(8, std::mem::size_of::<Option<llq::Node<AudioNodeId>>>());
        assert_eq!(16, std::mem::size_of::<Box<dyn AudioProcessor>>());
        assert_eq!(24, std::mem::size_of::<Vec<AudioRenderQuantum>>()); // times 2
        assert_eq!(8, std::mem::size_of::<ChannelConfig>());
        assert_eq!(64, std::mem::size_of::<SmallVec<[OutgoingEdge; 2]>>());
        assert_eq!(1, std::mem::size_of::<bool>()); // times 3

This is expected to grow because we need to use more SmallVecs in #368