Support int32 and int64 in TF Lite

Adds support for the int32 and int64 data types to our TF Lite implementation

Update TF Lite backend to 2.4

A long time in coming update to our TF Lite code from the 1.13 version of that library to 2.4. All previously tested models built in 1.13 and 1.15 continue to work with the update, and it should support models built with newer version of Tensorflow.

Support for scalar inputs and for boosted tree models

Tree Models require TensorIOTensorFlow ~> 2.0.8

Clean up xcconfig

Removed gratuitous linker flag in search for fix to #192

Remove saved model code

Address #49 and removes local copies of SavedModel APIs from TensorFlow support

See https://github.com/doc-ai/tensorio-ios/releases/tag/1.2.1 for more detail

Last version with SavedModel code

Locking TensorIOTensorFlow dependency to 2.0.1, which is the last version of that dependency which does not include the SavedModel APIs in its build.

Will be immediately followed by a 2.0.2 update without the SavedModel code which depends on a version of TensorIOTensorFlow that includes it.

Unfortunately I'm not versioning this pair of updates semantically because TensorIOTensorFlow is tracking the major.minor version of the TensorFlow build it vends.

I would have preferred to make this fix earlier, in which case none of the 1.1 or 1.2 updates to TensorIO would have been necessary and a consumer could have simply specified which version of TensorIOTensorFlow it wanted for the backend.

Compatibility updates for TF 2.0 backend

Updates Tensor/IO to work with the TF 2.0 build of our TensorFlow backend.

This backend has been tested for inference and training with models built up to the latest version of TensorFlow at the time of release, TF 2.2.

For more information on TF builds and Tensor/IO compatibility see TensorFlow Compatibility.

Compatibility updates for TF 1.15 backend

Updates Tensor/IO to work with the TF 1.15 build on the TensorFlow backend.

From this point on updates to Tensor/IO that are TF 1.15 compatible will be patch updates to this 1.1 release. Updates to Tensor/IO that are TF 1.13 compatible will be patch updates to the earlier 1.0 release.

We are working on building TF 2.0 and incorporating it into Tensor/IO, in which case the next release will target that build and backend.

For more information on TF builds and Tensor/IO compatibility see TensorFlow Compatibility

Locks TensorFlow backend to TensorIOTensorFlow 1.13

Ensures that all 1.0.x versions of Tensor/IO use the TF 1.13 backend moving forward. The next major release of Tensor/IO, v1.1 will use the TF 1.15 backend instead.

We made it!

Tensor/IO 1.0 is doc.ai's on device machine learning framework. Tensor/IO abstracts the work of interacting with an underlying ML library such as TensorFlow and provides a declarative interface to your machine learning models. Tensor/IO currently supports the TF Lite and TensorFlow backends and is extensible to other on device libraries.

For full TensorFlow models, Tensor/IO uses a custom build of TensorFlow v1.13 for iOS. This build supports models built in TF 1.x and some models built in TF 2. We are working on upgrading our on device build to TF 1.15, which should increase the number of models we support.

Moved documentation to https://github.com/doc-ai/tensorio/tree/master/documentation
Moved examples to https://github.com/doc-ai/tensorio/tree/master/examples
Moved all outstanding issues to 1.x milestone