Consider switching model to centerface_bnmerged.onnx

[mdraw]

Judging from the file name, https://github.com/Star-Clouds/CenterFace/blob/master/models/onnx/centerface_bnmerged.onnx is probably a model that is equivalent to the currently included one, with batch normalization ops merged into the adjacent convolution layers - a cheap and straightforward optimization for inference. Replacing the current model with the "bnmerged" version, deface runs a few percent faster than earlier in preliminary tests (CPU execution with opencv and onnxruntime backends).

TODO:

• [x] Measure performance improvement with longer running tests
  • Neglegible impact on OpenCV-based inference but ONNX-Runtime was about 10% to 20% faster in most tests.
• [x] Verify that results are comparable.
• [x] If above results are positive, replace the model file, update the credits section in the readme and make a new release.

[mdraw]

Comparing the two onnx models in https://netron.app/ the network structures indeed look the same (even including node IDs, so we don't have to do any code changes), except for the eliminated batch norm layers in the "bnmerged" model.