Add "replace" to streaming challenge with new runbooks

[magdalendobson]

This PR adds a new operation, "replace," to the options in the streaming benchmark challenge. It reflects the operation where an embedding for a particular vector ID is updated, with use cases such as computing a new embedding to reflect small changes to a document ID. The changes to the core repository are fairly small: the running and loading code now accept replace as an option, and it is included in the streaming groundtruth calculation script. All these changes are fully backwards-compatible with the original neurips23 runners.

We also contribute three new runbooks that use the new "replace" operation. The first is simple_replace_runbook.yaml, which is modeled after the existing simple runbook. The second, random_replace_runbook.yaml is based on the msturing-10m-clustered dataset. It first inserts a prefix of the data in each cluster, then replaces part of that prefix with data from a randomly selected cluster. The third, clustered_replace_runbook.yaml is also based on msturing-10m-clustered, and it inserts a prefix of the data in each cluster, then replaces part of that prefix with reserved data points from the same cluster. The script for generating these runbooks is deterministic and is provided in gen_replace_runbooks.py.

[magdalendobson]

In response to suggestions @harsha-simhadri, I added a README that documents the semantics of replace. I also used this as an opportunity to start collecting a list of streaming runbooks with brief descriptions, as we now have ten runbooks.