Adapt API for the upcoming SK changes

[neon-sunset]

Given likely changes in memory connector API in SK, track the changes and consider how Transport/Index and PineconeClient methods need to be adjusted to minimize unnecessary work and memory usage caused by e.g. constructing an async enumerable only for it to call a "unary" style of method, or having SK to create a batch and then upsert/fetch vectors one by one over the batched Pinecone API.