diff-test: summarize changes per field (depth=2) in final output

Description

Extend diff-test's final summary line so that — when there are changes — it also reports a breakdown of changed fields aggregated at depth 2 of the JSON path.

Currently the output looks like:

Took: 15.082626564s - 0 changes in 0 of 14409 files (0.0%)

When changes exist, it would be useful to also see something like:

Field summary (files affected):
  doc.body          1234
  doc.flaws          456
  doc.summary         12
  ...

Aggregation rule: multiple changes within the same depth-2 subtree on a single page count as 1 for that page (e.g. several diffs under doc.flaws.* → doc.flaws += 1 for that file).

Motivation

Makes it easier to understand the shape of a regression at a glance, without opening the HTML/CSV report.

Implementation notes

• The per-file diff in full_diff already produces dotted JSON-path keys in a BTreeMap<String, String>.
• Add a shared counter map (DashMap<String, AtomicUsize>) next to same / total_changes.
• After each full_diff call (in both the --value HTML branch and the --csv branch), compute the HashSet of depth-2 prefixes from diff.keys() and increment the counter once per prefix.
• Print the sorted summary (descending by count) after the existing Took: … line.
• The non-value HTML branch doesn't call full_diff, so the summary should be gated on --value or --csv.
• Note that IGNORED_KEYS currently excludes doc.flaws, doc.modified, doc.popularity, etc., so those won't appear in the summary unless removed from that list.

Start in crates/diff-test/src/main.rs, tracing full_diff and the --value HTML and --csv branches where per-file diffs are processed. Add the depth-2 per-file aggregation and confirm that, for those modes and only when changes exist, the sorted field summary appears after the existing Took line.