v0.66.10

Fewer round trips for parallel reads

The HTTP client now groups model reads started together into a bounded request and removes duplicate reads within that group. Applications that load several collections in parallel make fewer HTTP requests without changing their calls. Each read retains its own permissions, pagination, claims and transaction evidence; results are not cached between requests. Older servers continue to receive individual reads. The reduction takes effect after updating the SDK against a server that supports the batch endpoint.