Misleading "checkpoint" message when reusing dataset

[dmpetrov]

Description

DataChain currently uses the term "checkpoint" in user-facing messages for two different concepts:

1. recovery from incomplete execution (UDF-level checkpoints)
2. reuse of already computed datasets (save-level)

This leads to misleading UX. There is no problem with UDF-checkpoints (1) but dataset reuse leads to a misleading message:

Checkpoint found for dataset 'my_ds', skipping creation

However, "checkpoint" implies partial or incomplete progress and recovery failure. In this case, user thinks that failure happened in the last run.

It should be replaced to:

Reusing dataset 'my_ds@1.0.2' (skipping recompute)

This issue is about UX but this problem might affect internals as well.

[ilongin]

Agree, and this ties directly into the discussion we're having in #1653 about removing transient hash dependencies — see the checkpoint reuse scope analysis.

With transient dependencies (current main), checkpoints form a chain — each checkpoint hash includes the previous one (H1 → H2 → H3), marking sequential progress through the script. The term "checkpoint" makes sense here because it's literally checkpointing progress in a sequence.

Without transient dependencies (#1653), each chain's hash is calculated independently. There's no chaining. On re-run, each chain independently checks "was this computation done before?" and skips if yes. That's not checkpointing — it's memoization/caching.

Breaking it down by type:

	With transient deps	Without transient deps
UDF partial (crash recovery)	Real checkpoint	Real checkpoint
UDF final (skip UDF on re-run)	Checkpoint in a chain (links to next)	Cache/memoization
Save-level (skip dataset creation)	Checkpoint in a chain (links to next)	Cache/memoization

Only UDF partial recovery remains a true "checkpoint" after removing transient deps — saving intermediate progress during a long-running UDF so we can resume after a crash. Everything else becomes "has this been computed before? skip it."

[shcheklein]

The term "checkpoint" makes sense here because it's literally checkpointing progress in a sequence.

Tbh, I don't think so, even in this case. This ticket is just about UX that is more clear to users. And no matter what implementation is - we still have two types - full "save" that can be reused and partial save. And I think it makes sense to distinguish them in UX (and probably terminology). Full save it kinda an additional feature?