fix(storage): ignore duplicate large-file candidates

[rorschachwhy]

Summary

• make advisory native large-file candidate ingestion idempotent when a filesystem index repeats a path
• preserve strict duplicate rejection for authoritative traversal streams
• avoid double-counting progress, allocated/logical bytes, and file totals

Problem

On macOS, a full-disk large-file scan can fail while consuming Spotlight results with:

failed to consume the candidate stream: the in-memory index received a duplicate large-file record

Spotlight is an advisory candidate source and can repeat a path. A repeated, already-validated candidate should not abort the entire scan.

Safety

Live scope, protection policy, filesystem, link-type, and size validation still runs before insertion. Only the advisory candidate insertion is idempotent; the existing strict duplicate checks for authoritative traversal/index records remain unchanged.

Validation

• pnpm check
• cargo test --manifest-path src-tauri/Cargo.toml -p mangodisk-core (495 passed, 0 failed)
• deterministic regression test covering duplicate native candidates and aggregate accounting
• real macOS root-volume Spotlight fast-path scan completed successfully
• Windows was not run locally

[harry0703]

Thank you for identifying and fixing this issue. The distinction between advisory candidate ingestion and authoritative index validation is clear and well scoped.

I also verified the change on Windows ARM64: the duplicate-candidate regression test passed, the full Core suite passed with 523 tests and no failures, and pnpm check completed successfully. This improves scan reliability and keeps file-count and size accounting accurate across platforms.

Thanks again for the thoughtful contribution!

[harry0703]

Your fix is now included in 1.0.9. Repeated paths from Spotlight no longer interrupt large-file scans or cause files and sizes to be counted twice.

Thanks for the contribution and the regression test! I've also credited you in the release notes.