Replace vision hotswap with native multimodal profiles

dev/test_local.sh

@@ -124,7 +124,7 @@ import json,sys
 d=json.load(sys.stdin)
 ids={m["id"]:m for m in d["data"]}
 assert set(ids)=={"qwen-fast","qwen-medium","qwen-long"}, ids
-assert ids["qwen-fast"]["context_length"]==73728
+assert ids["qwen-fast"]["context_length"]==76800
 assert ids["qwen-medium"]["context_length"]==94208
 assert ids["qwen-long"]["context_length"]==131072
 ' && ok "drei virtuelle Modelle mit korrekten Context Windows" || bad "/v1/models"
@@ -138,7 +138,7 @@ import json,sys
 d=json.load(sys.stdin)
 assert d["current_profile"]=="fast", d
 assert d["upstream"]["reachable"] is True, d
-assert d["upstream"]["ctx"]==73728, d
+assert d["upstream"]["ctx"]==76800, d
 ' && ok "Status zeigt Profil fast + erreichbares Upstream" || bad "/status"
 
 # --- 3. Normales Forwarding (non-streaming) -------------------------------------
@@ -199,7 +199,7 @@ echo "$RESP" | python3 -m json.tool
 echo "$RESP" | python3 -c '
 import json,sys
 d=json.load(sys.stdin)
-assert d["profile"]=="fast" and d["context_length"]==73728, d
+assert d["profile"]=="fast" and d["context_length"]==76800, d
 ' && ok "Profil fast wieder aktiv" || bad "Profilwechsel fast"
 
 # --- 8. Virtuelles Modell triggert Profilwechsel ----------------------------------------
@@ -270,7 +270,7 @@ echo "$RESP" | python3 -m json.tool
 echo "$RESP" | python3 -c '
 import json,sys
 d=json.load(sys.stdin)
-assert d["profile"]=="fast" and d["model"]=="mock-model-73728", d
+assert d["profile"]=="fast" and d["model"]=="mock-model-76800", d
 ' && ok "Recovery: /fast wartet auf Upstream, dann Erfolg" || bad "Recovery"
 
 # --- 11. Bildgenerierung (Mock-Worker) -------------------------------------------------