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/text_norm.go

@@ -0,0 +1,15 @@
+package ucca
+
+import "strings"
+
+// normalizeGerman lowercases and folds German umlauts / ß to their ASCII digraphs
+// (ä→ae, ö→oe, ü→ue, ß→ss) so keyword matching is insensitive to whether the user
+// typed "Prüfe" or "Pruefe", "Datenschutzerklärung" or "Datenschutzerklaerung".
+// Applied to BOTH the query and the keyword lists in the German-text matchers.
+func normalizeGerman(s string) string {
+	return umlautFolder.Replace(strings.ToLower(s))
+}
+
+var umlautFolder = strings.NewReplacer(
+	"ä", "ae", "ö", "oe", "ü", "ue", "ß", "ss",
+)