Configurable resources for RGB and NN

[iamsg95]

New options were added:
sh/shaves - number of shaves used by NN
cmx/cmx_slices -number of cmx slices used by NN
NCE/NCEs -number of NCEs used by NN
rgb/rgb_cam_config - rgb camera config, 1080p_30hz / 4k_30hz supported currently

Install requirements for cloud compile:
python3 -m pip install -r requirements.txt
Install dependencies for hostapp cpp backend:

Ubuntu:
        sudo apt-get install -y  libcurl4-openssl-dev
macOS:
        brew install curl

Run as :
./depthai.py -s metaout previewout -sh 8 -cmx 8 -NCE 1 -rgb 4k_30hz -v test.h264 -dd

Tested on Mac and Linux.

[Luxonis-Brandon]

Thanks good idea.

[iamsg95]

Added model downloader into cpp backend:
New dependencies to build hostapp:

Ubuntu:
        sudo apt-get install -y  libcurl4-openssl-dev
  - macOS:
        brew install curl

[iamsg95]

Fixed @Luxonis-Brandon
I was about to move NN decodings in different file, so it's cleaner, so I added a new parameter "labels", but I forget to update the code at one place.

[Luxonis-Brandon]

Thanks @szabi-luxonis . Yes, tested it and I can confirm it looks great! No problem now on my Mac for the same command.

[Luxonis-Brandon]

All LGTM and works for me!

[alex-luxonis]

With a configuration that overloads the device so it can't keep up with 30fps processing, I see frame tearing or out-of-order frames, seems to appear in both previewout and h.264 streams, so could be from the ISP/post-processed output.

./depthai.py -sh 8 -cmx 8 -NCE 2 -rgbr 2160 -dd -v test.h264
./depthai.py -sh 8 -cmx 8 -NCE 2 -rgbr 2160 -dd

Not sure if caused by the change in frame update policy.
If not easy to resolve now, we could just flag the issue, as this may be an unusual configuration.

[Luxonis-Brandon]

Yes agreed on if not easy to solve to just flag it as I agree it is likely an unusual configuration.