Introduce cheaply clonable ChunkStoreHandle and QueryCacheHandle

[teh-cmc]

These are required for:

1. Making it easy to construct and pass QueryEngines around:

// TODO(cmc): This needs to be a refcounted type that can be easily be passed around: the ref has
// got to go. But for that we need to generally introduce `ChunkStoreHandle` and `QueryCacheHandle`
// first, and this is not as straightforward as it seems.
pub struct QueryEngine<'a> {
    pub store: &'a ChunkStore,
    pub cache: &'a Caches,
}

2. Getting ingestion off of the main thread and parallelize it:

• Parallelize data ingestion #4298

[teh-cmc]

@zehiko just hit that one as he was trying to integrate a QueryHandle/QueryEngine into an async Stream (which assumes multi-threaded environment, and therefore static lifetimes, etc).