fix(query-orchestrator): QueryCache - propagate error later for streams

[viktordebulat]

If the source (rowStream) emits an error or closes before all data is read, pipeline will reject.

This manually consumes the async iterator from the stream, catching errors at each step. If the stream ends early or throws, you catch the error and can handle it gracefully. This avoids some of the internal state issues that can occur with pipeline when the source stream is an async generator that may throw.

#9566

[viktordebulat]

Hi @igorlukanin, i'm not sure that this approach is the best way. Could you summon someone to review this? I've been using this for 3 month in my project (ClickHouse rollups with lambdas) and would like to add this fix to the upstream

[KSDaemon]

@ovr WDYT?