Improve the performance of the main functions in the library and benchmark them

[dsyme]

Repo Assist do this when you've not much else to do?

[github-actions]

🤖 This is an automated response from Repo Assist.

I've submitted a draft PR addressing this. The three highest-impact combinators that were still using the asyncSeq computation builder (or an indirect composition) have been rewritten with optimised direct enumerators:

• filterAsync — asyncSeq { for … } builder → OptimizedFilterAsyncEnumerator (direct while-loop, same pattern as mapAsync)
• chooseAsync (non-AsyncSeqOp path) — likewise
• foldAsync (non-AsyncSeqOp path) — scanAsync + lastOrDefault composition → direct loop, eliminating the entire intermediate async sequence

Two new BenchmarkDotNet benchmark classes (AsyncSeqFilterChooseFoldBenchmarks, AsyncSeqPipelineBenchmarks) are included to measure these operations and catch future regressions.

All 317 existing tests pass. The PR is filed as a draft for review.

Generated by Repo Assist · ◷

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@346204513ecfa08b81566450d7d599556807389f