Implement RecordStream for iterating over scanned records

[snowsignal]

This PR solves two problems with Recordset:

1. Iteration/collection of scanned records is currently not workable in an async context because the Recordset iterator calls std::thread::yield_now() internally.
2. An Arc<Recordset> can be sent between threads, and iterated over in multiple places at once, which would leave each thread with a partial iteration over the scanned records. Ideally, it should be impossible to iterate over the scanned records in more than one thread at once.

To solve these, I've implemented a new struct, RecordStream, that acts as an iterable wrapper over an Arc<Recordset>. It can be created by a new into_stream() method on Arc<Recordset>, and crucially, this will make at most one RecordStream available at any one time—trying to instantiate a second RecordStream will return None.

RecordStream implements both the synchronous Iterator trait that was previously implemented for &Recordset, but it also supports asynchronous iteration via futures::Stream.

It might be worth considering whether Client::scan should just return a RecordStream directly, though this can be done in a follow-up PR.

[khaf]

Thank you for your feedback and PR. I need some clarification though:

1. We have tests for scan and query that are running in a async context and clearly working. What am I missing? (the std::thread::yield_now() was added to aid sync client in consuming the records. If found to be an issue, I'll need to find another solution.)
2. If by partial iteration you mean a multi-consumer setup, then that is by design. Or do you mean the iterator gets stuck?

Is possible to provide a gist to help me understand and reproduce the issue on my side? Thank you.