There is no single approach. It depends on your organization.
First you have the 'impact-axis' which is lean and mean vs big bang.
Next you also have bottom-up (first fix bronze and work upwards) or top-down (focus on gold and read data from your legacy EDW using connections).
So a lot is possible.
I am always a fan on focusing on a single stream/cube/datamart and start migrating that.
That is more a leand and mean/bottom up approach, which i prefer.  But you might think otherwise.

Just don't think that your Databricks Lakehouse will not change.  With planning you can limit the amount of reengineering but not avoid it. Basically it is the same as a EDW but way less strict, which has its pros and cons.