Implementing TableProvider::schema and RecordBatchStream::schema(): knowing the schema before the stream is executed

[jwimberl]

Context: an implementer of TableProvider must implement the method fn schema(&self) -> SchemaRef, besides implementing the scan method that returns an ExecutionPlan implementation. And that ExecutionPlan implementation can return a wrapped RecordBatchStream implementation from its execute method, but that in turn must define a fn schema(&self) -> SchemaRef method as well, beyond implmenting Stream<Item = Result<RecordBatch, DataFusionError>> trait.

There are, though, conceivable circumstances where the only real way to know the schema is run through enough of the logic to get the first RecordBatch, or just shy of that some IPC header information, and then repeat the same in work in the eventual RecordBatchStream. Is there any strategy to avoid that double effort?