[SeriesDFilter] Redundant execution of filter_pipe for each fields_group causes slow data loading

[CanvaChen]

Description

During data loading, I noticed that filter_pipe is executed repeatedly for each fields_group instead of being applied once or cached. This causes significant slowdowns in data preparation, especially when multiple field groups are configured.

Steps to Reproduce / Observation

1. Added logging/print statements inside the _getFilterSeries method.
2. Ran the data loading pipeline with multiple fields_groups (e.g., 3 groups in my setup).
3. Observed that the filter pipeline was called exactly 3 times, once per fields_group, rather than sharing the filtered result across groups.

[he-yufeng]

I found a small loader-side fix for this: resolve ilter_pipe once at the QlibDataLoader.load() level, then reuse the filtered instruments across field groups. I added a regression test that checks D.instruments(..., filter_pipe=...) is called once while grouped feature loading still runs per group. I'll open the PR shortly.