cli: output polish batch from the friction walk

[mbertschler]

Problem (friction log F7, F8, F11, F18, F19, F24 — S3/S4)

Checklist of small output fixes from the reference-setup walk; none needs design work:

• F7 — transferred=0 is ambiguous: sync summaries never report the already-correct count, so a no-op-because-in-sync is indistinguishable from a no-op-because-empty. Add already_correct=N (CLI summary + peer-sync report line).
• F8 — indexing an empty/mistyped volume path is silent: added=0 … errors=0 on an empty directory looks identical to a wrong mount. Warn when an index run finds a volume path empty ("new volume or wrong mount?"), at minimum on first index.
• F11a — index/sync summary lines don't name the volume/pair (running several back-to-back yields anonymous lines).
• F11b — rclone.conf updated prints even for kopia-only syncs.
• F11c — kopia's ANSI color codes leak escape sequences into the structured agent log.
• F11d — refused syncs print status= … run=0 (empty status, zero id) — superseded by runs: every failure and refusal becomes a run row or a standing state #157's refused run rows, but the line rendering should cope either way.
• F18 — inbound and outbound peer runs are indistinguishable in runs/TUI (destination column holds a peer name in both directions). Label direction (e.g. → htpc vs ← laptop).
• F19 — steady-state no-op rows drown signal: runs --failed / --changes filters; TUI folding of consecutive no-op rows. (Also: runs --help still says "List index runs"; it lists all kinds.)
• F24 — catch-up moments deserve a summary: after a burst run (many files after quiet), the summary could say so ("404 new files after 21 days") instead of blending into the noise. Cheap heuristic, big trust payoff.
• Restore summary prints the sync-shaped arrow (docs → cloudbox) even though bytes flowed destination→local; flip the arrow for restores.

Each item is independently landable; fine to split across small PRs referencing this issue.

See design/friction-log.md for the walk context per finding.

[mbertschler]

Landed by #179, which merged into the claude/issue-157-run-rows-standing-states branch and reached main with #174 — the stacked base is why the Closes #161 keyword never auto-closed this.

Verified on main (09c2760): every checklist item is in — already_correct=N (F7), the empty-volume index warning (F8), named summary lines (F11a), no rclone preamble for kopia-only syncs (F11b), ANSI stripping on kopia stderr (F11c), suppressed counter line for pre-runs-row failures (F11d), peer-direction arrows in runs and the TUI (F18), runs --failed / --changes plus the TUI f fold and the corrected help text (F19), the catch-up note (F24), and the flipped restore arrow.

One deliberate partial from F19 is now tracked separately in #182: --changes and the fold key on file_count == 0, which collapses peer-sync no-ops but leaves bucket/index no-ops visible, because runs.file_count counts files considered for those kinds. A real changed-file count needs a v27 migration, out of scope while #157/#158 owned v25/v26.

Closing as completed.

---

Generated by Claude Code