perf(ai-sdk): embed query once across router fan-out + fold umlauts in intent/concept matching

Authority Router re-embedded the query per collection (6x); on dev the embed
endpoint (OVH) is remote so that was 6 round-trips = 7-12s per /retrieve. Embed
once, reuse via ctx across the concurrent per-collection searches.
DetectIntent + ConceptNorms now fold ae/oe/ue/ss so ASCII (Pruefe) and umlaut
(Pruefe) inputs both match.

ai-compliance-sdk/internal/ucca/umlaut_test.go

@@ -0,0 +1,29 @@
+package ucca
+
+import "testing"
+
+func TestDetectIntentUmlautFold(t *testing.T) {
+	cases := map[string]string{
+		"Pruefe meine Datenschutzerklaerung.": "review",     // ASCII digraph
+		"Prüfe meine Datenschutzerklärung.":   "review",     // umlaut
+		"Ueberpruefe das Impressum":           "review",     // ASCII "überprüfe"
+		"Was ist eine TOM?":                   "definition", // unchanged
+	}
+	for q, want := range cases {
+		if got := DetectIntent(q); got != want {
+			t.Errorf("DetectIntent(%q)=%q want %q", q, got, want)
+		}
+	}
+}
+
+func TestConceptNormsUmlautFold(t *testing.T) {
+	// ASCII "datenschutzerklaerung" must resolve to the same core norms as the umlaut form.
+	ascii := ConceptNorms("Was gehoert in eine Datenschutzerklaerung?")
+	umlaut := ConceptNorms("Was gehört in eine Datenschutzerklärung?")
+	if len(ascii) == 0 {
+		t.Errorf("ConceptNorms(ASCII datenschutzerklaerung) returned none")
+	}
+	if len(ascii) != len(umlaut) {
+		t.Errorf("ASCII vs umlaut concept norms differ: %v vs %v", ascii, umlaut)
+	}
+}