Improvements around C++ code + better error handling + two new OpenCV functions + clearBuffers support for skipping some objects (by ids)

[experiment322]

This is a big diff that I've accumulated while integrating this project into one of my projects. I'm not sure about the changes in build.gradle, especially the one pinning the react-native version, but that helped me get gradle and Android Studio to work and to help with C++ code while working on this.

Trying to resume the changes:

• added calcOpticalFlowPyrLK and phaseCorrelate OpenCV functions
• fixed checks for optional arguments inside the invoke C++ function (check argument count before asserting the type, because otherwise we're accessing out of bounds indexes in the arguments array -> bad memory access?)
• also implemented the index check in FOCV_FunctionArguments and throw some errors if the index is out of bounds + also implemented a type check before trying to cast (if the index check passed) + implemented some new OpenCV types (I think TermCriteria is the only one?)
• fixed FOCV_Object Mat creation when given an array with existing data (it didn't respect the number of channels specified in the requested type, always reshaping to 1) + improved array value extraction performance (reserved the required length ahead of time) + again added some index checks for vectors - other small fixes?
• implemented a whitelist of ids to keep when clearing buffers in FOCV_Storage::clear so you can exclude some objects from the cleanup because you might need them later (I used this lib in combination with react-native-vision-camera, and inside the frame processor - a function that runs for each frame from the camera - I did some opencv stuff and I needed to retain an image for the next loop while clearing the other stuff to avoid filling the memory and to improve performance) - the implementation is questionable as it accepts an array of ids due to simplicity, so in js you'd have to pass [opencvObject.id] instead of [opencvObject]
• maybe other stuff?

I opened the PR because I considered it's a nice improvement that would benefit everybody, but it's not very refined, well tested and very well documented yet, so any feedback is welcome. I might revert some changes, especially from the build.gradle, but they don't seem to break anything yet. Also, these changes shouldn't introduce any breaking change. Also, I didn't test very much on iOS, but will do in the near future.

PS: Thank you for this library!

PS PS: calcOpticalFlowPyrLK works pretty well, that's the one I'm mainly using, but phaseCorrelate I'm not so sure of, I tested it a little and it was outputting some values, but they seemed strange to me, but maybe I didn't fully understand it. I was wondering if I'm doing some bad casting, but not sure.

[lukaszkurantdev]

@experiment322 Wow, thanks for your work! Sorry for the delay, but I need some more time to process it. I'll add it to the next version, but I need to do some tests and improvements in different parts.

[lukaszkurantdev]

@experiment322 Thanks one more time for your improvements. I've merged them to current lib repo.

Unfortunately I need to remove calcOpticalFlowPyrLK support – it's function from object detection module of OpenCV and currently it's not supported on iOS (library currently uses a slimmed-down customised build of OpenCV without this module).

Other changes remain and will be adapted for a future version :)

[experiment322]

Ok, thanks for getting this in! But I have two things to mention:

• I was and I'm still not sure if the build.gradle changes are good, but if it doesn't break anything I suppose it's fine
• I didn't document the storage selective clear functionality, you might want to do that when you got time so people are aware of it