feat: add basic explain get_metadata function for tf2.

[ji-yaqi]

Basic explain get_metadata function, for tf2.

Usage:

builder = google.cloud.aiplatform.explain.metadata.tf.v2.saved_model_metadata_builder.SavedModelMetadataBuilder(
    "MODEL_PATH"
)
metadata = builder.get_metadata()

The main difference from Vertex SDK and XAI SDK's get_metadata function:

1. In Vertex SDK, users the camelCase while in XAI, variables users snake_case
   e.g. Vertex: inputTensorName, XAI: input_tensor_name

2. If modality and encoding in inputs are not set, XAI SDK populates the default:

      'modality': 'categorical',
      'encoding': 'identity'

Vertex SDK doesn't populate the default values. It only populates the users's settings.
The two methods will not make any difference in the actual settings.

[ji-yaqi]

/hold

[ji-yaqi]

/unhold

[sasha-gitg]

This Optional type is a List while the default is an empty tuple. It seems like the default can be None instead.

[sasha-gitg]

Option to reduce the size of this block.

Suggested change

	if tensor_spec.dtype.is_floating:
	input_mds[
	name
	] = explanation_metadata.ExplanationMetadata.InputMetadata(
	input_tensor_name=name
	)
	else:
	input_mds[
	name
	] = explanation_metadata.ExplanationMetadata.InputMetadata(
	input_tensor_name=name, modality="categorical",
	)
	input_mds[name] = explanation_metadata.ExplanationMetadata.InputMetadata(
	input_tensor_name=name,
	modality=None if tensor_spec.dtype.is_floating else "categorical"
	)

[sasha-gitg]

"tensorflow" is the CPU-only release: https://www.tensorflow.org/install/pip

This will also make it easier to manage with TB's TF dependency.

[sasha-gitg]

tensorboard and xai have the same dependencies so we may need to deduplicate this with list(set(...)).

[sasha-gitg]

Since this is only used by one test, it should perhaps not be a shared method.