Remove the need to Box rcl_node_t

[mxgrey]

I was able to recreate the crash that happens in humble, as discussed here. It does seem that in Humble rcl is doing some questionable things with the rcl_node_t address.

This PR offers an alternative solution that does not require us to Box the rcl_node_t. Instead we initialize the rcl_node_t while it is already positioned inside the NodeHandle. Since the NodeHandle has a fixed memory address throughout its lifetime, the rcl_node_t will also have a fixed address.

The only downside is we need to add another variable to NodeHandle to track whether the rcl_node_t was successfully initialized, otherwise it might attempt to call rcl_node_fini on an invalid rcl_node_t instance which lead to bad and confusing behavior in the tests. I think overall this is less disruptive than adding a layer of indirection to rcl_node_t.

[geoff-imdex]

That's a much nicer fix than using a Box.