The interesting question here is how to register the message types for the range values when converting a table schema to the descriptor proto. At a high level, there's three options:

• publish a set of messages similar to the well-known types, and only add the reference. This makes compilation more awkward, as these types aren't general enough to warrant include in the common protobuf libraries, nor even in the google API type extensions.

• [current implementation] generate type names independently of where they occur in the message hierarchy. This makes it easier to recognize a range message via the TypeName, since we lose the ability to recognize the RANGE purely from the descriptor.

• inject range messages into the common dependency cache as we're walking the referenced messages. This consolidates the dependency cache a bit, but it means the TypeName of the message will be based on the first encountered reference (something like root.nested_message.range_field). It also means that any user messages with a similar two-field schema (start/end) could potentially be re-used in the cache.