Fixed marshalling of size_t into nuint instead of int.

[p-lindberg]

This PR fixes the issue of marshalling size_t into appropriately sized integers, as described in this issue.

I went with the native nuint type, which should map 1-to-1 with size_t.

For the public API, I decided to cast the nuint returned by mdb_cursor_count to a long, as this seems more idiomatic and convenient for users when dealing with counts. However, this is a breaking change, and if avoiding it is preferred then we could cast to an int instead (but see below). I do think it makes sense to return a long though, as (I assume) the underlying implementation supports having more than 2^32 duplicate entries in a database, in which case an int would truncate the value.

For the transaction id, on the other hand, I decided to propagate the nuint to the public API. This seems like the safest option as it makes no assumptions on the possible values that can be returned by the C implementation. In principle one could probably cast safely to a long, but as an identifier it isn't meant to be used in arithmetic operations anyhow. Either way, I think this needs to be a breaking change, as an int transaction id is unsafe on platforms where the underlying implementation can return values greater than 2^32 - this can lead to id reuse, which could become a problem for long-running processes.