"CREATE TABLE (REPLACE TABLE) ... AS SELECT" Support

[sungwy]

Feature Request / Improvement

In Spark SQL, we have the ability to combine the table overwrite using AS SELECT statement, with create_table or replace_table, as an atomic operation. (CTAS, RTAS)

Do we intend to support this feature with the same atomicity guarantee in PyIceberg?

Since the PyIceberg client is in charge of writing out the manifests and constructing the new table metadata, I think it is technically possible. Would we just add

as_select: pa.Table = None

as an optional parameter to create_table and replace_table and add a snapshot update with full table static overwrite to the new table metadata?

[Fokko]

Duplicate of #281

[sungwy]

Duplicate of #281

This was my bad attempt at decoupling the introduction of REPALCE TABLE support from the discussion of how we should support ... AS SELECT semantic - but I think it's a bit too late for that now given how I've described the previous issue #281. I will bring these points back to the other issue.