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

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

internal/proxy/hosts.go

@@ -75,7 +75,7 @@ func (c *hostChooser) Choose(candidates []string, model string) (string, bool) {
 			Draining: false,
 			Loaded:   contains(s.Loaded, model),
 			CanServe: c.cfg.Serves(host, model),
-			Free:     c.lim.FreeSlots(host),
+			Free:     freeForModel(c.cfg.Hosts[host], model, c.lim.InFlight(host, model)),
 			Queued:   c.lim.Queued(host, model),
 			Weight:   c.cfg.Hosts[host].Weight,
 		}
@@ -91,3 +91,17 @@ func contains(list []string, v string) bool {
 	}
 	return false
 }
+
+// freeForModel returns the free slots for one model on one host: its parallel minus the in-flight
+// count, floored at zero, and zero when the host does not list that model.
+func freeForModel(h config.Host, model string, inflight int) int {
+	m, ok := h.Models[model]
+	if !ok {
+		return 0
+	}
+	free := m.Parallel - inflight
+	if free < 0 {
+		return 0
+	}
+	return free
+}