feat(wing): score a wing by its best page, not summed hits; add wingbacktest command (#205 S1)

A first backtest showed BM25 summed over a wing's top hits favours big wings
(0/4 agreement). Best-page scoring is volume-robust; pinned by a test that
failed before the change.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

ingestion/internal/wing/bm25scorer_test.go

@@ -43,3 +43,19 @@ func TestBM25Scorer_NoUsableTermsScoresNothing(t *testing.T) {
 	s := wing.NewBM25Scorer(scorerBrain(t), []string{"infra", "finance"}, "")
 	assert.Empty(t, s.Score("a an to of"))
 }
+
+// A big wing must not win just by having more pages that brush the query: a
+// wing is scored by its single best page, not by the sum of its top hits.
+func TestBM25Scorer_ABigWingDoesNotWinOnVolume(t *testing.T) {
+	dir := t.TempDir()
+	writePage(t, dir, "wiki/small/facts/target.md", "---\ntitle: Target\n---\nkubernetes cluster scheduling pods deployment nodes\n")
+	for i := 0; i < 8; i++ {
+		writePage(t, dir, filepath.Join("wiki/big/facts", "p"+string(rune('a'+i))+".md"),
+			"---\ntitle: Filler\n---\nkubernetes cluster scheduling mentioned in passing among bookkeeping ledger entries\n")
+	}
+	s := wing.NewBM25Scorer(dir, []string{"small", "big"}, "")
+	got := s.Score("kubernetes cluster scheduling pods deployment nodes")
+	top, ok := wing.TopWing(got)
+	assert.True(t, ok)
+	assert.Equal(t, "small", top)
+}