Document the mining pipeline and the LFS/Xet fetch paths

[khatchad]

The README documents only the three original mining scripts and still describes the project as the 2025 GSTEM study. It does not mention analysis/inspect_commit.py, the candidate set at data/message_refactoring_candidates.csv, or how the LFS and Xet fetch paths behave, so a new contributor cannot get from a clone to an inspected commit without asking.

Much of what makes the inspection path work is currently unwritten, and it took several rounds to get right. That knowledge should live in the repo rather than in anyones head.

What to cover

• What analysis/inspect_commit.py does, and its flags.
• When --download is required, and what it actually fetches (the header row rather than the whole file).
• When --show_rows is needed, and why a column-set diff alone is insufficient for a values-only change. See --show_rows diffs the local LFS pointer instead of the downloaded content #52.
• Which formats are read, and which produce a "not analyzed" notice. See Read modified TSV and gzipped tables, and name the files left unread #59.
• The failure modes that look like success. A confident empty result is not evidence that nothing changed, and that shape has caused real mislabelling. See inspect_commit.py --download: crash on LFS files, and a silent "no column change" when the pull does not materialize #34.
• The LFS and Xet fetch paths, including anything non-obvious about how they are selected and what goes wrong. See Inspect column changes for LFS-tracked CSVs #23 and did issue #30 #32.

Provenance section

Update it so the current state of the project is represented, rather than only the 2025 preliminary study. The classifier spec lives in #11.