Considering that they had to rework things to not use the master key, I can understand the time and effort. From TFA, the DB Gateway handled all Cosmos requests, and used the master key to get an account’s private key and then forward the DB requests. If they had to rearchitect that gateway system, I’m sure it was no small task. Bear in mind, they probably approached this somewhat seriously since Cosmos powers lots of internal M$ tools like Teams and Copilot, not just customer applications.
They patched the vulnerability within 2 days. They spent 6 months rearchitecting their backend. It's a major service for themselves and tens of thousands of customers, you don't vibe code yourself a new database query execution engine overnight.