@Xiphware and I further discussed in Slack and I've given this further thought as I wrote more about dynamic defaults in repeats.
I think it boils down to whether we believe that the initialization phase as defined by W3C XForms is something that happens exactly once per instance or every time an instance is loaded. To be explicit, I'm not talking about primary vs. secondary instances here, I'm talking about a specific record built from the primary instance which has a unique instanceID (in ODK XForms, any secondary instances are treated as read-only and there is only support for exactly one model).
xforms-ready is a notification-only event fired right after xforms-model-construct-done triggers any of its actions so since we are only proposing to support exactly one of these, my understanding is that they would be identical. Since we have no authoritative way of confirming the intent of these events, I continue thinking that defining our own events with clear semantics makes the most sense. If for some reason we do one day need to support more of the W3C XForms events, that path will still be open and we'll just have to be very explicit about the order in which they are fired relative to the custom ODK XForms events.
Additionally, odk-instance-first-load has now shipped and is in use. My sense is that for a change to be worth it we'd either have to decide that the mechanism we've used is bad in some way or that an alternative would have user-facing benefits.