Your message seems to mostly be concerned about issues during row inserts and
updates. I agree that this is most important to get correct, and my observation
does not suggest any problems with this.

The issue I am concerned about is only related to table creation, not data
modification. I think it is a bad idea to allow the creation of a table that
cannot be created on a different system with different libz. This could cause
problems in many scenarios (like replication eg.)

Hence the suggestion to use the same, most conservative (but still accurate)
upper bound when deciding whether the CREATE TABLE is valid. The libz
documentation provides such an upper bound, as I wrote.