Smoke run for v1; README for leases, admin and accounting

Implemented-By: OpenCode session (model recorded in docs/implementer-log.md)

cmd/fakeupstream/main.go

@@ -1,11 +1,13 @@
 // fakeupstream stands in for a llama-server router in tests and the smoke run. Do not edit.
 //
-//	fakeupstream -listen 127.0.0.1:18081 -name alpha -models a,b -down-file /tmp/alpha.down
+//	fakeupstream -listen 127.0.0.1:18081 -name alpha -models a,b -down-file /tmp/alpha.down -slow 0
 //
 // /health answers 503 while the down file exists, 200 otherwise. /v1/models lists -models.
 // /props answers a small JSON object. /v1/chat/completions echoes: a streamed answer of five
-// SSE chunks 200 ms apart when the body has "stream": true, one JSON answer otherwise. Every
-// response carries X-Upstream: <name>.
+// SSE chunks 200 ms apart when the body has "stream": true, then a final chunk carrying
+// "usage" and llama-server style "timings", then [DONE]; one JSON answer with usage and
+// timings otherwise. -slow adds that many milliseconds before answering (for queue tests).
+// Every response carries X-Upstream: <name>.
 package main
 
 import (
@@ -25,11 +27,14 @@ func main() {
 	name := flag.String("name", "fake", "name reported in X-Upstream and answers")
 	models := flag.String("models", "m", "comma-separated model ids for /v1/models")
 	downFile := flag.String("down-file", "", "while this file exists, /health answers 503")
+	slow := flag.Int("slow", 0, "milliseconds to wait before answering a completion")
 	flag.Parse()
 
 	ids := strings.Split(*models, ",")
 	mux := http.NewServeMux()
 	stamp := func(w http.ResponseWriter) { w.Header().Set("X-Upstream", *name) }
+	usage := map[string]any{"prompt_tokens": 100, "completion_tokens": 10, "total_tokens": 110}
+	timings := map[string]any{"prompt_n": 100, "cache_n": 90, "predicted_n": 10, "predicted_ms": 50.0}
 
 	mux.HandleFunc("/health", func(w http.ResponseWriter, r *http.Request) {
 		stamp(w)
@@ -61,11 +66,12 @@ func main() {
 			Stream bool   `json:"stream"`
 		}
 		_ = json.Unmarshal(body, &req)
+		time.Sleep(time.Duration(*slow) * time.Millisecond)
 		if !req.Stream {
 			writeJSON(w, map[string]any{
 				"id": "chatcmpl-fake", "object": "chat.completion", "model": req.Model,
 				"choices": []map[string]any{{"index": 0, "message": map[string]string{"role": "assistant", "content": "hello from " + *name}, "finish_reason": "stop"}},
-				"usage":   map[string]int{"prompt_tokens": 3, "completion_tokens": 3, "total_tokens": 6},
+				"usage":   usage, "timings": timings,
 			})
 			return
 		}
@@ -73,16 +79,24 @@ func main() {
 		w.Header().Set("Cache-Control", "no-cache")
 		w.WriteHeader(http.StatusOK)
 		fl, _ := w.(http.Flusher)
+		flush := func() {
+			if fl != nil {
+				fl.Flush()
+			}
+		}
 		for i := 1; i <= 5; i++ {
 			chunk := map[string]any{"id": "chatcmpl-fake", "object": "chat.completion.chunk", "model": req.Model,
 				"choices": []map[string]any{{"index": 0, "delta": map[string]string{"content": fmt.Sprintf("%s chunk %d ", *name, i)}}}}
 			b, _ := json.Marshal(chunk)
 			fmt.Fprintf(w, "data: %s\n\n", b)
-			if fl != nil {
-				fl.Flush()
-			}
+			flush()
 			time.Sleep(200 * time.Millisecond)
 		}
+		final := map[string]any{"id": "chatcmpl-fake", "object": "chat.completion.chunk", "model": req.Model,
+			"choices": []map[string]any{}, "usage": usage, "timings": timings}
+		b, _ := json.Marshal(final)
+		fmt.Fprintf(w, "data: %s\n\n", b)
+		flush()
 		fmt.Fprint(w, "data: [DONE]\n\n")
 	})
 	mux.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
@@ -90,7 +104,7 @@ func main() {
 		http.Error(w, `{"error":"not found"}`, http.StatusNotFound)
 	})
 
-	log.Printf("fakeupstream %s listening on %s models=%v", *name, *listen, ids)
+	log.Printf("fakeupstream %s listening on %s models=%v slow=%dms", *name, *listen, ids, *slow)
 	srv := &http.Server{Addr: *listen, Handler: mux, ReadHeaderTimeout: 5 * time.Second}
 	log.Fatal(srv.ListenAndServe())
 }