fix(io): return correct list_status paths and reuse storage operators

[QuakeWang]

Purpose

Fix FileIO::list_status path semantics: each returned FileStatus.path now points to the actual listed entry instead of the input directory path.
Also, keep reusable operators in Storage to preserve memory backend state across calls.

Brief change log

• Fix list_status to return real entry paths (base_path + entry.path()), and fetch required metadata via list_with(...).metakey(...).
• Refactor Storage to reuse Operator instances across calls (including memory backend).
• Add regression tests for list_status on both fs and memory backends.

Tests

• cargo fmt --all -- --check
• cargo clippy --all-targets --workspace -- -D warnings
• cargo test -p paimon io::file_io::file_action_test
• cargo test --all-targets --workspace

All tests passed locally.

API and Format

• No public API signature changes.
• No storage format changes.
• Behavior fix only: list_status now returns correct entry paths.

Documentation

No documentation update required.

[QuakeWang]

@Aitozi @JingsongLi PTAL 👀

[JingsongLi]

cc @luoyuxia to take a review

[luoyuxia]

@QuakeWang Thanks for the pr. Left minor comments. PTAL

[luoyuxia]

IIUC, format!("{base_path}{}", entry.path()) just add leading / to entry.path()?
Considering openal will always call normalize_path to remove leading /, do we real need to add eading / to entry.path()? Could you just keep entry.path()?

[QuakeWang]

@luoyuxia Thanks for the comment. entry.path() in OpenDAL is relative to operator root, while FileStatus.path here is expected to preserve the original FileIO path prefix (e.g., memory:/ or
file:/) for consistency with get_status and follow-up FileIO operations. normalize_path applies to Operator input paths, not to Entry::path() output.

[luoyuxia]

Thanks. It sounds reasonable to me to return the full path with scheme prefix.

But it doesn't seem to be necessary now for reusing the storage operator. Maybe you could create a separate PR for reusing the storage operator part in which we can revist it further it.

[luoyuxia]

Left minor comments.

[luoyuxia]

Thanks. It sounds reasonable to me to return the full path with scheme prefix.

But it doesn't seem to be necessary now for reusing the storage operator. Maybe you could create a separate PR for reusing the storage operator part in which we can revist it further it.

[QuakeWang]

@luoyuxia Thanks for the suggestions. I narrowed this PR to the list_status path fix only and reverted the storage-operator reuse changes. I also simplified the new test by removing non-essential assertions and reducing duplicated write code. The FS path regression test is kept.

[luoyuxia]

@QuakeWang Thanks. +1

[JingsongLi]

+1