Improve dataset upload retries for completed and incomplete files

### Task Summary

Improve dataset upload retry behavior so batch uploads distinguish between incomplete multipart uploads and files that already exist in the dataset.

| Case | Expected behavior |
| --- | --- |
| Active multipart upload session exists for the same path | Prompt the user to resume or restart the incomplete upload. |
| A file with the same path and size already exists in committed or staged dataset files | Prompt the user to upload again or skip the matching file. |

The completed-file prompt should use cautious wording because matching by path and size does not prove byte-for-byte equality.

Implementation should include:

- A backend dataset-scoped check for candidate upload paths and sizes.
- Frontend logic that checks active multipart sessions first, then checks existing matching files.
- Support for mixed retry batches where one file resumes and another file can be skipped.
- Tests for multipart resume behavior, completed-file skip behavior, backend committed/staged matches, and invalid or unauthorized requests.

Related discussion: #5744
Related PR: #5929

### Task Type

- [ ] Refactor / Cleanup
- [ ] DevOps / Deployment / CI
- [x] Testing / QA
- [ ] Documentation
- [ ] Performance
- [x] Other