fix+feat: offene Punkte abgeschlossen

Kontext in Connect-Snippet:
- ConnectPanel: max_tokens nutzt jetzt den echten konfigurierten Kontext
  des Modells (m.meta.ctx) statt hartkodiertem 32768 -> Zed sieht 128k

Phase D - 'Immer aktiv halten' Toggle:
- models.py: UpdateReq bekommt optionales ttl-Feld
- update_model: setzt TTL wenn angegeben (ctx und ttl unabhaengig)
- ModelsPanel: Checkbox 'Immer aktiv halten' im Konfig-Modal
  (TTL=99999 = nie entladen, TTL=300 = Standard 5min)

MoE tps-Schaetzung:
- hw_math.py: extract_active_params_b() erkennt A3B-Suffix
- estimate_speed(): moe_active_ratio-Parameter, sqrt-Boost fuer MoE
- evaluate_fit(): name-Parameter (optional) fuer automatische MoE-Erkennung
- cookbook.py: alle evaluate_fit-Aufrufe mit name= versehen

Cleanup:
- static/js/panels/*.js + static/js/main.js geloescht (Dead Code)
- wird-Datei (versehentlich committet) geloescht
- CLAUDE.md: KISS-Statement auf Vite+Svelte-Stand aktualisiert

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

routers/cookbook.py

@@ -127,7 +127,7 @@ async def analyze_repo(req: AnalyzeRequest):
     results = []
     for f in gguf_files:
         quant = extract_quant(f)
-        fit = evaluate_fit(params_b, quant, req.ctx, ram_gb)
+        fit = evaluate_fit(params_b, quant, req.ctx, ram_gb, name=req.repo_id)
         
         # Priority-Score, um den besten Fit an oberste Stelle zu setzen.
         # "Q4_K_M" ist oft der Sweetspot.
@@ -173,7 +173,7 @@ def recipes():
     for r in all_recipes():
         models, worst = [], "perfect"
         for m in r["models"]:
-            fit = evaluate_fit(m["params_b"], m["quant"], 8192, ram_gb)
+            fit = evaluate_fit(m["params_b"], m["quant"], 8192, ram_gb, name=m.get("name", ""))
             models.append({**m, "fit": fit, "optimal_ctx": max_ctx_for(m["params_b"], m["quant"], ram_gb)})
             if _FIT_ORDER[fit["level"]] > _FIT_ORDER[worst]:
                 worst = fit["level"]
@@ -427,7 +427,7 @@ def refresh_discover(ram_gb: float) -> dict:
         role = _categorize(rid)
         params_b = extract_params_b(rid)
         quant = "Q4_K_M"  # Referenz-Quant fuer die Fit-Einschaetzung
-        fit = evaluate_fit(params_b, quant, 8192, ram_gb)
+        fit = evaluate_fit(params_b, quant, 8192, ram_gb, name=rid)
         by_cat[role].append({
             "name": rid.split("/")[-1], "author": rid.split("/")[0], "repo": rid,
             "role": role, "params_b": params_b, "quant": quant,
@@ -518,7 +518,7 @@ def compute_upgrades(ram_gb):
         out.append({
             "name": up["name"], "repo": up["repo"], "params_b": up["params_b"], "quant": up["quant"],
             "why": up["why"], "old": old, "role": old,
-            "fit": evaluate_fit(up["params_b"], up["quant"], 8192, ram_gb),
+            "fit": evaluate_fit(up["params_b"], up["quant"], 8192, ram_gb, name=up.get("name", "")),
             "optimal_ctx": max_ctx_for(up["params_b"], up["quant"], ram_gb),
         })
     return out

routers/models.py

@@ -58,7 +58,8 @@ class ChatReq(BaseModel):
 
 class UpdateReq(BaseModel):
     alias: str
-    ctx: int
+    ctx: int | None = None
+    ttl: int | None = None
 
 
 class DeleteReq(BaseModel):
@@ -251,17 +252,20 @@ def update_model(req: UpdateReq):
     cfg = read_config()
     if req.alias not in cfg.get("models", {}):
         raise HTTPException(404, "Modell nicht gefunden")
-    
+
     spec = cfg["models"][req.alias]
-    cmd = str(spec.get("cmd", ""))
-    
-    # Replace or add context size
-    if re.search(r'-(?:c|-ctx-size)\s+\d+', cmd):
-        cmd = re.sub(r'-(?:c|-ctx-size)\s+\d+', f'-c {req.ctx}', cmd)
-    else:
-        cmd = cmd.strip() + f" -c {req.ctx}\n"
-        
-    cfg["models"][req.alias]["cmd"] = LiteralScalarString(cmd)
+
+    if req.ctx is not None:
+        cmd = str(spec.get("cmd", ""))
+        if re.search(r'-(?:c|-ctx-size)\s+\d+', cmd):
+            cmd = re.sub(r'-(?:c|-ctx-size)\s+\d+', f'-c {req.ctx}', cmd)
+        else:
+            cmd = cmd.strip() + f" -c {req.ctx}\n"
+        cfg["models"][req.alias]["cmd"] = LiteralScalarString(cmd)
+
+    if req.ttl is not None:
+        cfg["models"][req.alias]["ttl"] = req.ttl
+
     write_config(cfg)
     return {"ok": True}