fix: cache filtered instruments during grouped loads

[he-yufeng]

Description

QlibDataLoader.load() now resolves filter_pipe once before grouped field loading starts, then reuses the filtered instruments for each field group. Direct load_group_df() calls keep the old behavior by resolving instruments at that entry point.

A regression test covers grouped loading with two field groups and verifies that D.instruments(..., filter_pipe=...) is called once while D.features(...) is still called per group.

Motivation and Context

Fixes #2236.

The previous grouped path delegated every group to load_group_df() with the original market name, so each group re-ran the same dynamic instrument filter pipeline. With multiple fields_groups, that multiplies filter work even though the filtered instrument set is identical for the load request.

How Has This Been Tested?

• Pass the test by running: pytest qlib/tests/test_all_pipeline.py under upper directory of qlib.
• If you are adding a new feature, test on your own test scripts.

Targeted checks run locally:

• python -m py_compile qlib\data\dataset\loader.py tests\data_mid_layer_tests\test_dataloader.py
• git diff --check
• an isolated loader smoke test that stubs Qlib's C extensions and asserts one D.instruments call plus two grouped D.features calls

I also attempted the targeted pytest:

• python -m pytest tests\data_mid_layer_tests\test_dataloader.py::TestDataLoader::test_group_loader_applies_filter_pipe_once -q

That is blocked in my Windows checkout before test collection because the local Cython extension is not built:

ModuleNotFoundError: No module named 'qlib.data._libs.rolling'

Building the extension with python setup.py build_ext --inplace also fails locally because this Windows environment cannot find the Windows SDK io.h header while compiling rolling.cpp.

Screenshots of Test Results (if appropriate):

1. Pipeline test: not run; local C extension build is blocked as described above.
2. Your own tests: py_compile, diff check, and isolated loader smoke test passed.

Types of changes

• Fix bugs
• Add new feature
• Update documentation