Add a hand_landmarker.task for Greyscale images 8 bits

[AndrewBloom]

MediaPipe Solution (you are using)

mediapipe/tasks/c/vision/hand_landmarker

Programming language

C

Are you willing to contribute it

None

Describe the feature and the current behaviour/state

The current implementation does not support ImageFormat::GRAY8 format

Will this change the current API? How?

No

Who will benefit with this feature?

Everyone using it

Please specify the use cases for this feature

Using grayscale images will reduce the number of calculation necessary given the same amount of pixels. Moreover the grayscale image is immediately accessible from YUV888 without a costly conversion to RGB format first

Any Other info

No response

[AndrewBloom]

Basically running it on Android platform all on NDK side, using the C API.
Perfectly working if I convert the YUV888 to RGB format, but not if I pass directly the buffer YUV and select ImageFormat::GRAY8.
Error is:
INVALID_ARGUMENT: Graph has errors:
Calculator::Process() for node "mediapipe_tasks_vision_hand_landmarker_handlandmarkergraph__mediapipe_tasks_vision_hand_detector_handdetectorgraph__mediapipe_tasks_core_inferencesubgraph__inferencecalculator__mediapipe_tasks_vision_hand_landmarker_handlandmarkergraph__mediapipe_tasks_vision_hand_detector_handdetectorgraph__mediapipe_tasks_core_inferencesubgraph__InferenceCalculator" failed: ; RET_CHECK failure (mediapipe/calculators/tensor/inference_calculator_utils.cc:144) (tflite_tensor.bytes)==(input_tensor.bytes())TfLiteTensor and Tensor sizes do not match. TfLiteTensor dims: [1, 192, 192, 3], type: FLOAT32, bytes: 442368 vs. MP Tensor dims: [1, 192, 192, 1], type: kFloat32, bytes: 147456; at index 0 [MediaPipeTasksStatus='606']

Main advantages are:
Skipping the Image conversion, Less operations necessary given the same image.
Are color informations adding anything anyway?