Remove refcounting from reply/sample.

[jean-roland]

Oudated.

_For Zenoh-C compatibility issue, we have a single z_loaned_sample_t type which corresponds both to a sample received by a sub and the sample contained in a reply.

This type is refcounted as required for potential out-of-context usage by a subscriber. As replies are themselves refcounted that create a double refcount which is slightly wasteful memory-wise in a Query/reply scenario.

This PR attemps to offer a way to avoid this double refcounting by storing the reply sample without refcounting, provide a zenoh pico api to interact with it, while a call to the Zenoh-C api will still generate the recounted sample._

[jean-roland]

Alternative solution after discussion with @sashacmc and @DenisBiryukov91: Make the allocated fields (payload, attachments...) refcounted and make the containers (reply, sample) copyable rather than refcounted.

To avoid doubling interfaces.

[DenisBiryukov91]

I just realized, that query likely should stay ref counted, because zenoh allows to make copies of queries and send multiple replies through them. The query is considered to be finalized when the last copy is dropped.