Trtutils

Added
- `impls.yolo.ParallelYOLO`: Enables running multiple YOLO models simultaneously.

Improved
- `TRTEngine`:
- Uses async memory copies and execution.
- Implements pagelocked memory on host.

Removed
- `backend` submodule: Deprecated in favor of CUDA Python engines.

---

Added
- `jetson.benchmark_engine` integrated with `jetsontools > 0.0.3`.

Improved
- `TRTEngine`: Enhanced threading documentation.

Fixed
- `trtexec.build_engine`: Correctly builds for DLA core 0.

---

Changed
- `TRTEngine`:
- Uses `execute_async_v2` for inference.
- `core.create_engine` now creates a `cudaStream`.

---

Added
- Locks for TensorRT engine creation and CUDA memory allocation.

---

Added
- `benchmark_engine`: Measures engine latency.
- Submodules:
- `jetson`
- `impls`
- `impls.yolo`: Supports YOLO variants (V7 to V10).

Changed
- `trtexec.build_from_onnx` renamed to `trtexec.build_engine`.

---

Added
- Async and parallel execution classes:
- `QueuedTRTEngine`, `QueuedTRTModel`
- `ParallelTRTEngine`, `ParallelTRTModel`

---