[EPIC] Stateful transactions and snapshot production (RFC-0003) #3266

What's the feature are you trying to implement?

This epic tracks the implementation of the stateful transaction architecture defined in RFC-0003: Stateful Transactions and Snapshot Production (introduced in #2620).

Today, transaction actions are replayed from scratch on every catalog conflict. That is fine for simple metadata updates, but snapshot-producing actions (appends, and eventually merging operations like RowDelta/RewriteFiles) benefit from retaining work across retry attempts: stable execution identity, reusable metadata processing, and ownership of generated files that may require cleanup.

The RFC establishes:

• A stateful TransactionAction model: each action pairs immutable intent with exclusively-owned, retry-persistent state; the transaction owns catalog refresh, retry, and replay.
• Persistent snapshot producers (SimpleSnapshotProducer, MergingSnapshotProducer) as action state, replacing the attempt-scoped producer and its hook traits.
• Artifact ownership and terminal cleanup semantics (Committed / Failed / Unknown).
• Reuse rules for retained work across retry/rebase.

Milestones (per RFC §8.1)

• 1. Stateful actions and replay — associated State type, action-entry ownership, fresh-state cloning; migrate existing actions with State = () (feat(transaction): Introduce stateful TransactionAction trait #3265)
• See milestones 2-4 in sub issues

Out of scope

End-to-end operation implementations built on top of this foundation — e.g. RowDelta / RewriteFiles — are tracked separately (#1607) and become unblocked once milestones 2–4 land.

Related issues

• Add RewriteFiles support to iceberg-rust #1607 — RewriteFiles support (builds on milestones 2–4)
• Process delete files when writing snapshots #2242 — process delete files when writing snapshots
• Support validate and related logic in Snapshot #1344 / Add validate_added_delete_files #2517 — snapshot validation logic (addressed by milestone 4)
• Unify RewriteManifestsAction with SnapshotProducer once Replace/Delete ops are supported #2556 — unify RewriteManifestsAction with snapshot production
• SnapshotProducer can't count removed files or rewrite a manifest under its own schema #3252 — SnapshotProducer manifest-rewrite limitations (superseded by the producer rework)

Willingness to contribute

I can contribute to this feature independently