materialize-s3-iceberg: s/python/iceberg-go/

[jacobmarble]

Description:

Replaces the embedded Python iceberg-ctl subprocess with the native Go iceberg-go library. The connector now creates and writes to Iceberg tables in-process.

New tables are created with metadata retention defaults (gzip metadata, drop prior versions on commit, expire snapshots older than 24h) to keep metadata files under object-store size limits.

As a nice side-effect, this also works around a date range issue in Python. When Iceberg metadata contains dates outside of the years 1 through 9999, it fails to parse because the corresponding Python type has this limit. Now that Python is removed from the code path, these dates are parsed without issue.

Workflow steps:

No user-visible changes.

Documentation links affected:

None.

Notes for reviewers:

• iceberg-ctl/ Python tree is removed wholesale.
• go/schema-gen/generate.go now traverses oneOf/anyOf branches so fixups apply to nullable-wrapped fields emitted by invopop.

[mdibaiee]

@jacobmarble do we still want to do this?
I'm just afraid of the risk of regressions with this, seems like a sudden lift, but I like the path... mixing Python and Go is quite the mess

[jacobmarble]

@jacobmarble do we still want to do this? I'm just afraid of the risk of regressions with this, seems like a sudden lift, but I like the path... mixing Python and Go is quite the mess

I still want to do it, sooner is better IMHO.

The Python Iceberg implementation is more mature than the Golang impl, but we don't push it very hard.

The PR hasn't been updated recently because no one had reviewed it yet. If you support the effort (with caveat to mitigate regression risks) then I'll rebase and squash to make the review easier.

[mdibaiee]

@jacobmarble yeah I think this is worth doing if we can mitigate the risk 👍🏽