[ROSETTA] Detect same-size binary changes during publish

[isolomatov-gd]

In src/rosetta-cli/rosetta_cli/services/document_data.py, non-text content contributes only its byte length to content_hash, so different binary files of the same size are treated as unchanged and skipped during publish. Hash the raw bytes or a deterministic byte digest for binary files, and add an equal-size replacement regression test.

[github-actions]

Rosetta Triage

Classification: bug — the change detector can suppress a publish after binary content changes.
Priority assessment: Medium. It affects binary assets only, but causes stale published content without a visible error.

Findings:

• Confirmed: DocumentData.from_file uses the byte count, rather than the byte content, as the hash input for non-text files.
• Confirmed: ContentPublisher._has_content_changed_cached compares that value with stored metadata and skips a matching document.
• Existing document-data tests cover text and metadata hash changes only; they do not cover binary content or same-size replacement.

Suggestions:

• Include raw binary bytes, or a deterministic digest of those bytes, in the content-hash input while preserving the existing metadata fields.
• Add a regression test that replaces a binary file with different bytes of the same length and verifies its content hash changes; an integration-level assertion should also verify publish does not skip it.

Next steps: Update binary hash construction and add the regression coverage.

Automated triage by Rosetta agent