Related issue(s)/PRs:

Summary

Adds gpflow trajectory sampling for simple additive kernels.

Note: this temporarily points to a new gpflux branch. This will of course be removed before merging, and once a gpflux release is made. Related GPflux PR: secondmind-labs/GPflux#104.

As an additional test, I ran the scalable_thompson_sampling_using_sparse_gaussian_processes.pct.py notebook with the model kernel replaced with an additive kernel (diff below), and it ran fine.

@@ -57,6 +58,8 @@ from trieste.models.optimizer import BatchOptimizer
 gpflow_model = build_svgp(
     initial_data, search_space, likelihood_variance=0.01, num_inducing_points=50
 )
+# Create a simple additive kernel.
+gpflow_model.kernel = Sum([RBF(), gpflow_model.kernel]) 

Fully backwards compatible: yes

PR checklist

• The quality checks are all passing
• The bug case / new feature is covered by tests
• Any new features are well-documented (in docstrings or notebooks)