fix request parsing

cpp/request_parsing.cpp

@@ -56,8 +56,8 @@ std::optional<TokenEstimate> estimate_chat_tokens(
     }
 
     if (estimate.requested_max_tokens <= 0) {
-        error = "'max_tokens' must be > 0";
-        return std::nullopt;
+        // -1 (and other non-positive values) is a sentinel meaning "unlimited" — use the default
+        estimate.requested_max_tokens = limits.default_max_tokens;
     }
 
     for (const auto &message : payload["messages"]) {