The write-generated-reference-tables shouldn't overwrite tables by default

[MrPowers]

When I run write-generated-reference-tables, I'll normally only want to add the new reference tables that are missing. I won't want to overwrite the existing reference tables. Would be cool to not overwrite by default and add a flag that needs to be set if you want them to all be overwritten.

[edmondop]

One fundamental challenge here is that maybe some code to generate the table has been modified, and therefore you really want to regenerate existing tables.

We can leave the option to the user to decide, or write the parquet data to a tmp folder and then read the two tables and compare to make the process idempotent, or also publish in a json file more information about the data which will be generated (but it only works for small data)