Opencv-python-inference-engine

*inference time measured for tests on my laptop with i7-8550U*

I'll attach both wheels (JIT and OpenBLAS) to the release, but will upload to PyPI the OpenBLAS version only.

Changelog

1. New versions of all packages
+ OpenCV-4.5.1
+ dldt-2021.2
+ ffmpeg-4.3.2
+ OpenBLAS-0.3.13
2. New versioning (see README)
3. New branch without OpenBLAS (because JIT and OpenBLAS GEMMs now how comparable speed)
4. Examples added
5. Check Ubuntu version before compile anything added
6. Clean dldt tmp files between build added
7. Changed LICENSE files in python package to Apache2, as Intel did

OpenBlas

| GEMM | Inference time (ms) | Package size (MB) |
|--------------------------|---------------------------|---------------------------|
|JIT | 39.5 | 40.1 |

+ inference-engine aka dldt aka openvino updated -- 2021.4
+ ffmpeg updated -- n4.4
+ speed test removed due to problems with downloading a testing model from the google drive
+ text recognition example -- minor updates

v2021-04-13
+ New InferenceEngine version -- 2021.3 and OpenCV -- 4.5.2
+ Major test refactoring (less dependencies, clearer code)
+ InferenceEngine (dldt) cmake changed, now it compiles faster and have less size
+ Main version now has `-D GEMM=JIT`, because inference speeds of JIT and OpenBLAS are the same
+ OpenBLAS branch remain for historical reasons

New versions of opencv and dldt

OpenCV-4.5.0
dldt-2021.1
ffmpeg-4.3.1
OpenBLAS-0.3.10

Due to my mistake, `opencv_python_inference_engine-4.5.0.0` requires GLIBC_2.29 (I built it on Ubuntu 20.04).
The `opencv_python_inference_engine-4.5.0.1` is an Ubuntu 18.04 build.

New versions of opencv, dldt and ffmpeg
+ OpenCV-4.4.0
+ dldt-2020.4
+ ffmpeg-4.3.1
+ OpenBLAS-0.3.10

New versions of dldt, ffmpeg and OpenBLAS

+ OpenCV-4.3.0
+ dldt-2020.3
+ ffmpeg-4.3
+ OpenBLAS-0.3.10