[WIP] FFI Implementation

[craiglabenz]

Summary

Adds a MediaPipe implementation that talks directly to C++ via FFI. (Technically, indirectly and through a C intermediary, but still fairly directly in comparison to method channels.)

C bridge code uses Dart's native FFI functionality and ffigen to automatically generate the required bindings from C headers pulled from the MediaPipe repository.

TODOs

Note: This is a feature / release branch, which means only fully reviewed changes are able to land here via their own PRs.

• Add mediapipe_core package (google/flutter-mediapipe#11)
• Add mediapipe_text package (google/flutter-mediapipe#12)
  • Vendor MediaPipe libraries (google/flutter-mediapipe#9)
• Automate header file updates (google/flutter-mediapipe#10)
• Add mediapipe_vision package (google/flutter-mediapipe#13)
• Explore web implementation (which will use JS interop instead of FFI)

TESTING

To run this PR locally, complete the following steps (on macOS):

1. In a checkout of google/mediapipe with the latest internal changes pulled down, run this build command:

bazel build --config darwin_arm64 --strip always -c opt --define MEDIAPIPE_DISABLE_GPU=1 mediapipe/tasks/c/text/text_classifier:libtext_classifier.dylib

2. Copy the resulting libtext_classifier.dylib file (found in the bazel-out directory) to this repository at the packages/mediapipe-task-text/examples/assets directory.

Note: This will move up a level to mediapipe-task-text/assets after google/flutter-mediapipe#9.

3. Download bert_classifier.tflite and place it in the same directory.

4. Run the example app and click the floating action button to classify text.