feat: add incremental inference helper

[gyx09212214-prog]

Summary

Adds a lightweight helper for users who update their local qlib data themselves and only need predictions for newly available dates.

The new helper configures an existing DatasetH so the handler loading window includes the required historical warmup rows while the prediction segment stays limited to the incremental date range. For TSDatasetH, the warmup length is inferred from step_len - 1 unless hist_ref is provided explicitly.

Fixes #2073.

Changes

• Add get_inference_hist_ref, get_incremental_inference_config, and prepare_incremental_inference_dataset in qlib.workflow.online.update.
• Reuse the same incremental window helper inside DSBasedUpdater.prepare_data to keep the existing online updater behavior aligned with the public helper.
• Document the lightweight incremental prediction workflow in the online component docs.
• Add regression coverage for warmup-window calculation, dataset configuration, TSDatasetH warmup inference, and validation errors.

Validation

• python -m py_compile qlib\workflow\online\update.py tests\rolling_tests\test_incremental_inference.py
• git diff --check

I also attempted python -m pytest tests\rolling_tests\test_incremental_inference.py -q locally on Windows. Collection is blocked before the new tests run because the source checkout does not have the compiled qlib Cython extension (qlib.data._libs.rolling), and building it requires Microsoft Visual C++ Build Tools in this environment.