Documentation for pair_matching_model() predict() functionality is misleading

# 🐛 bug report

The documentation for pair_matching_model.predict() states that this updates clusters however this is incorrect - this API call only applies the current ML model to generate predictions for the current pairs.

## 🤔 Expected Behavior

Documentation should either remove the reference to predict (as it exists within MachineLearningModel section) or explain that calling predict() will apply the current ML model to generate labels for unverified pairs/records.

From MachineLearningModel:

> predict(**options)
> Suggest labels for unverified records.

## 😯 Current Behavior

Within the developer interface:
> pair_matching_model()
> Machine learning model for pair-matching for this Mastering project. Learns from verified labels and predicts categorization labels for unlabeled pairs.
>
> Calling predict() from this dataset will produce new (unpublished) clusters. These clusters are displayed on the “Clusters” page in the Tamr UI.

## 💁 Possible Solution

## 🔦 Context

The current documentation makes it difficult to understand what different calls are achieving in the standard mastering workflow.

## 💻 Code Sample

## 🌍 Your Environment

| Software | Version(s) |
| ----------------- | ---------- |
| tamr-unify-client | 0.9.0
| Tamr server |
| Python |
| Operating System |