feat(qec): add trt_decoder_config for real-time decoding

[wsttiger]

Add comprehensive configuration support for TensorRT decoder in the real-time decoding system.

Add TensorRT Decoder Configuration Support for Real-Time Decoding

Summary

This PR adds comprehensive configuration support for the TensorRT decoder in the real-time decoding system, enabling users to configure TRT decoders through the standard decoder_config interface with full YAML serialization support.

Motivation

The TensorRT decoder previously lacked integration with the real-time decoding configuration system. This made it impossible to configure TRT decoders using YAML files or the standard decoder_config API that other decoders (nv-qldpc, sliding_window, etc.) support.

Changes

1. Core Configuration Struct

Added trt_decoder_config struct in decoding_config.h with five parameters:

• onnx_load_path (string): Path to ONNX model file (builds TRT engine)
• engine_load_path (string): Path to pre-built TensorRT engine (faster startup)
• engine_save_path (string): Path to save built engine for reuse
• precision (string): Inference precision mode (fp16, bf16, int8, fp8, tf32, best)
• memory_workspace (size_t): Workspace memory size in bytes (default: 1GB)

2. Integration with Decoder Config System

• Added trt_decoder_config to the decoder_config variant
• Implemented to_heterogeneous_map() and from_heterogeneous_map() methods
• Added YAML serialization support via MappingTraits<trt_decoder_config>
• Integrated with set_decoder_custom_args_from_heterogeneous_map() for type dispatch

3. Python Bindings

• Exposed trt_decoder_config to Python via pybind11
• All fields are accessible as properties with proper type checking
• Exported as cudaq_qec.trt_decoder_config()

4. Comprehensive Tests

Added 3 test functions covering:

• Default value initialization (all fields default to None)
• Field operations: set, get, clear (parameterized across all 5 fields)
• YAML roundtrip: serialize to YAML and deserialize back

Usage Example

Python:
import cudaq_qec as qec

Create TRT decoder config

trt = qec.trt_decoder_config()
trt.engine_load_path = "/path/to/model.engine"
trt.precision = "fp16"

Use with decoder_config

dc = qec.decoder_config()
dc.type = "trt_decoder"
dc.set_decoder_custom_args(trt)

YAML roundtrip

yaml_str = dc.to_yaml_str()
dc2 = qec.decoder_config.from_yaml_str(yaml_str)C++:
using namespace cudaq::qec::decoding::config;

// Create TRT decoder config
trt_decoder_config trt;
trt.engine_load_path = "/path/to/model.engine";
trt.precision = "fp16";

// Use with decoder_config
decoder_config dc;
dc.type = "trt_decoder";
dc.decoder_custom_args = trt;

// YAML roundtrip
auto yaml_str = dc.to_yaml_str();
auto dc2 = decoder_config::from_yaml_str(yaml_str);### Testing

All tests pass: ✅ 34/34 in test_decoding_config.py

Test breakdown:

• 7 new TRT decoder config tests (3 functions, parameterized to 7 test cases)
• 27 existing tests continue to pass
• Coverage includes: initialization, field operations, type validation, YAML serialization

Run tests with:
pytest libs/qec/python/tests/test_decoding_config.py -k "trt_decoder" -v### Files Changed