fix: use RawDocumentBuf insteadof Document in MongoDB extractor

[caiq1nyu]

This change is motivated by two considerations:

• BSON deserialization is relatively expensive and CPU-intensive.
• Since Rust String must contain valid UTF-8, deserializing BSON with invalid UTF-8 content will fail.

To address these issues:

• Avoid BSON deserialization whenever possible, or deserialize only the minimal subset of fields that is required.
• Defer BSON parsing as much as possible, ideally until the sinker stage.

fix: #321