Fix UUID check for existing tables without a UUID

[rdblue]

This fixes two cases when assigning UUIDs to tables without them:

1. If the table is loaded, then assigned a UUID by a concurrent writer
2. If the table is written with a null UUID because no UUID was assigned

[rdblue]

@jzhuge FYI

[jzhuge]

+1 LGTM

A bit of context:

• This is to handle an metadata JSON file created before Add table UUID #264. JSON field table-uuid does not exist, thus metadata.uuid() returns null.
• If you still have pre Add table UUID #264 writer running (not recommended), you might have to disable the checkState temporarily.

[aokolnychyi]

I think this PR would also resolve the problem when users upgraded to an Iceberg version that started assigning table UUIDs but their first operation wasn't a child of SnapshotProducer (e.g. update table properties). Such operations would write an explicit null as the table UUID that will corrupt the old logic.

[aokolnychyi]

@rdblue, do we want to bring this in? There is only one open question on HadoopTableOperations.

[rdblue]

Yes, I just need to update it after your review. Sorry I'm taking so long getting back to it.

[rdblue]

@aokolnychyi, I updated HadoopTableOperations. Can you take another look at this?

[aokolnychyi]

+1

There is a relevant checkstyle violation. @rdblue, could you fix it and merge this PR?

[rdblue]

Merged. Thanks for the review, @aokolnychyi!