Change of Waitable waitset interface

Feature request

Feature description

Currently the rclcpp::Waitable operates directly on the waitset.

  add_to_wait_set(rcl_wait_set_t * wait_set) = 0;

  virtual
  bool
  is_ready(rcl_wait_set_t * wait_set) = 0;

The problem here is, that the executor is not directly aware of the changes to the waitset done
bye the Waitable. Therefore the executor can not do certain optimizations, like use precomputed waitsets.
Another drawback of the current implementation is, that we need to hold a shared_ptr to the waitable
while we call rcl::wait in order to make sure that the rcl primitives in the waitset do not get deleted.

Therefore I would propose to change the interface of the rclcpp::Waitable to something like this:

struct WaitableSetHandles
{
std::vector<std::shared_ptr<rcl_subscription_t> subscription_handles;
std::vector<std::shared_ptr<rcl_client_t> client_handles;
.
.
.
// If this triggers, the executer must request a new WaitableSetHandles from the waitable
std::optional<std::shared_ptr<rcl_guard_condition_t>> waitable_changed;
};

struct WaitableSetHandleState
{
std::std::vector<bool> subscription_is_ready;
std::std::vector<bool> client_is_ready;
.
.
.
};

class Waitable
{

/**
* Will be called before rcl::wait will be called. This may
* be used for rearming guard_conditions, if there is still
* data to be processed.
*/
virtual void before_wait_cb();

/**
* Returns all rcl handles used by this waitable
*/
virtual WaitableSetHandles getWaitsetHandles() = 0;

/**
* Returns all rcl handles used by this waitable
*/
virtual bool
  is_ready(WaitableSetHandleState * state) = 0;

};

Implementation considerations

This has an impact on a public interface, and though might/will break downstream implementations.
On the pro side, we can optimize a code path that is being heavily used reducing the cpu load and latency of the standard
executors.
Note, with this change, we could also integrate an optimization into the executor, that should get rid of most of the std::weak_ptr::lock() calls.

@wjwwood and @mjcarroll whats you opinion and this ?