Add regression test suite for datasets (QA)

[yuchen814]

What does this PR do?

Adds a data-driven manual/semi-automated regression test suite for datasets at tests/regression/regression_suite.md (128 test cases).

The suite was built by mining 984 historical issues (2020-04 → 2021-09) to find the bug categories that keep repeating, cross-checking the currently-open issues via the GitHub Search API, and mapping the result onto the documented QA focus areas. It defines a 30-minute smoke pass (24 P0 cases) and a 180-minute full regression run, each case with preconditions, numbered steps, and expected results, plus a traceability matrix back to specific issues.

Categories driving the most bugs (historical)

1. Data loading (48%) 2. Caching (29%) 3. Splits/slicing (29%) 4. Arrow serialization (26%)

Coverage gaps filled

• trust_remote_code (post-dates the historical snapshot; 36 open today)
• Streaming (low historical volume but high-growth) — deliberately over-weighted
• Multiprocessing hangs/freezes (e.g. save_to_disk() freezes when saving on s3 bucket with multiprocessing #6936, Allow passing a multiprocessing context to functions that support num_proc #6357)

Before submitting this PR, please make sure you have read the contribution guidelines.