Add combined, per-root, and comparison summary scopes

[helly25]

Add --summary-scope to combine statistics across roots, report each root separately, or summarize comparison categories. Existing summary groupings can now accompany comparison counts without losing left/right size information.

AG;DR

• Add all (default), root, left-only, right-only, different, and identical scopes. left/right alias the one-sided categories; diff and compare expand to ordered category lists. Expansion removes duplicates, and repeating the flag replaces the list.
• Scope selection is independent of per-path --compare-select. Paired categories emit separate tables for the two roots. Combined comparison statistics count both copies, including identical content.
• Preserve ordinary summary grouping, templates, archive accounting, collection/shard feeds, and hash-verification feeds. Category scopes retain per-entry contributions until comparison classifies them; they do not replay expressions or actions. Combined and root scopes aggregate during the walk.
• Keep --summary=compare as comparison counts and percentages. Bare --summary in comparison mode additionally requests ordinary counts/sizes when a scope is explicit. Scope without an active file-summary driver is a usage error, including --compare=summary --summary-scope=all and scopes following --summary=none. The shortcut expands to --summary=compare; users add --summary or a grouping for file statistics. Validation precedes traversal and actions.
• Share summary rendering, label explicit scope/root output, preserve default ordinary output, and emit zero totals for empty populations. JSONL exposes scope and root.
• Update registry help, comparison topic/examples, generated reference, changelog, and design documentation.

Validation: all seven affected Bazel targets passed (summary_test, globals_test, help_render_test, help_topic_test, run_test, xff_markdown_test, xff_notice_test). Tests include three roots, unequal pair sizes, category aliases/deduplication, combined counts, empty roots, replacement semantics, templates/collections, and invalid contexts. Normal pre-commit checks include changed-file clang-tidy, clang-format, shfmt, and shellcheck. macOS validation used the existing local Apple linker workaround; CI validates the supported platform matrix.

Category statistics require retained per-entry reduction data in addition to the comparison entries. Entries excluded from comparison classification do not enter category tables, even where special feeds include them in whole-root statistics. --summary-scope does not change histogram populations.

Follow-up validation: summary CLI and engine tests pass with regressions for both reported shortcut/scope combinations, explicit comparison-only summaries, clearing/restoring summaries, and rejection before execution. Regenerated Markdown passes its diff test; changed-file clang-tidy and formatting checks pass. Main now includes #831; its changelog-only conflict is resolved while retaining the feature notes.

Coverage follow-up: add direct engine and end-to-end cases for plain scope labels, count/name tie-breaking under --top, and empty comparison scope rejection. These exercise the 14 changed lines reported uncovered by CI. Both test targets pass locally; local coverage collection produced no usable line counters, so the measured patch percentage is verified by the Linux CI coverage job. No production behavior or thresholds changed in this follow-up.

CI coverage verification on 8aa6addaf: patch lines 275/275 (100.00%), branches 174/180 (96.67%), rating GOOD. The coverage gate passed.