Add a cv::dnn binding (Net class: readNet / setInput / forward)

[TroyHernandez]

Adds a binding for OpenCV's dnn module, as discussed in #101 (PR 1 of 2: dnn first, YuNet face detection to follow).

What

A Net class wrapping cv::dnn::Net:

• readNet(model, config = "", framework = "") loads a model (ONNX, Caffe, TF, Darknet, ...).
• net$setInput(image, scalefactor, size, mean, swapRB, crop, name) preprocesses an Image via blobFromImage and sets it as input.
• net$forward(name) runs a forward pass and returns the output tensor as a numeric array, with OpenCV's row-major layout converted to a correctly-shaped column-major R array of arbitrary dimensionality.
• net$getLayerNames(), net$empty(), and an isNet() helper round it out.

Notes

• No build change. dnn is already in ROpenCVLite's default module list, so this compiles against what's already there (no BUILD_LIST change, no contrib module, no new system dependency).
• Follows the existing conventions: one header per module (src/dnn.h), an RCPP_MODULE, and an R wrapper (R/dnn.R), mirroring the Image/Video classes.
• Tested locally against OpenCV 4.13: loads ONNX and TFLite models, runs forward, and returns correctly-shaped arrays.

Targets develop. Happy to adjust the API/conventions however you prefer.

🤖 Generated with Claude Code