runs: record a real changed-file count so `--changes` and the TUI fold cover bucket and index runs

[mbertschler]

Problem

runs.file_count means different things per run kind. For peer-sync runs it is
the number of files the sync actually acted on; for bucket (rclone) syncs and
index runs it is files considered, not moved. So there is no column that
answers "did this run change anything?" across kinds.

That left a deliberate gap in the F19 work from #161 (PR #179):

• runs --changes (runIsInteresting in cmd/squirrel/runs.go) and the TUI's
  f fold (runIsNoOp in tui/runs.go) both key on
  status == success && file_count == 0.
• That predicate collapses peer-sync no-ops correctly, but a bucket or index
  no-op has a non-zero file_count, so it stays visible — the deliberately
  fail-safe choice (better a visible no-op than a hidden run that moved
  content), documented in both call sites.

Result: in the reference household's steady state the filters and the fold
still leave the bulk of the noise on screen, because most rows are per-destination
bucket pushes. F19's "steady-state no-op rows drown signal" is only half solved.

Why it wasn't done in #161

It needs a schema migration, and at the time #157 owned v25 and #158 owned v26
(both now merged; SchemaVersion is 26). #179 was scoped as reporting-only —
no schema change — so the honest partial landed with the caveat recorded rather
than a silent half-fix.

Scope

• Add a real changed-file count to runs in a v27 migration — a new column
  (e.g. changed_count INTEGER, NULL for historical rows so old runs stay
  truthfully "unknown" rather than falsely "no-op"). Per the convention in
  AGENTS.md, any new table is STRICT; this is an ALTER … ADD COLUMN on
  the existing non-STRICT runs table, so it rides the current shape (see
  Adopt SQLite STRICT tables in the index schema #148 for the bulk conversion).
• Populate it from what each handler already knows: rclone's transferred
  count for bucket pushes (the same source as already_correct=N from F7),
  the disposition list for peer syncs, added/superseded for index runs.
• Switch runIsInteresting / runIsNoOp to the new column, falling back to
  today's file_count == 0 heuristic only where changed_count IS NULL, so
  pre-migration history keeps its current (conservative) rendering.
• Regenerate store/schema.sql (go test ./store -update-schema) and extend
  the runs filter tests to cover a bucket no-op collapsing and a bucket run
  that moved content staying visible.

Notes

• Purely additive: no existing column changes meaning, and no run row is ever
  rewritten — the audit trail stays intact.
• Once this lands, already_correct=N (F7) and the fold agree on one number,
  which is also what status: per volume×destination coverage + durability panel (CLI, then TUI in this issue) #159's status grid wants when it says "caught up".

Follow-up to #161 (F19); the caveat is spelled out in PR #179's description.