refactor: Remove maxNumSequences parameter from MakeDecodingBatchInputOutput

- Removed maxNumSequences parameter from createDecoderBatchInputs and related function calls, streamlining the interface.
- Updated all relevant implementations and tests to reflect the changes in function signatures.

Signed-off-by: Robin Kobus <[email protected]>

Author: Funatiq

cpp/tensorrt_llm/pybind/batch_manager/algorithms.cpp

@@ -136,7 +136,7 @@ void tensorrt_llm::pybind::batch_manager::algorithms::initBindings(pybind11::mod
         .def(py::init())
         .def("__call__", &MakeDecodingBatchInputOutput::operator(), py::arg("decoder_input_buffers"),
             py::arg("decoder_state"), py::arg("context_requests"), py::arg("generation_requests"), py::arg("logits"),
-            py::arg("model_config"), py::arg("max_num_sequences"), py::arg("fused_runtime_buffers") = std::nullopt)
+            py::arg("model_config"), py::arg("fused_runtime_buffers") = std::nullopt)
         .def("name", [](MakeDecodingBatchInputOutput const&) { return MakeDecodingBatchInputOutput::name; });
 
     py::class_<LogitsPostProcessor>(m, LogitsPostProcessor::name)