Fix: Remove unused attachment data from ipynb files when attachment links are deleted

[Copilot]

When users paste images into Jupyter notebook markdown cells and later delete the attachment references from the markdown content, the base64-encoded image data was still being saved in the .ipynb file. This caused unnecessary file bloat and potential data leakage.

Problem

The issue occurred in the serialization phase where createMarkdownCellFromNotebookCell() and createRawCellFromNotebookCell() functions blindly copied all attachments from cell metadata to the serialized output, regardless of whether those attachments were actually referenced in the cell content.

Solution

• Added getReferencedAttachmentNames() helper function that uses regex to detect actual attachment references in markdown content
• Modified serialization functions to filter attachments based on actual usage
• Only attachments that are referenced with ![...](attachment:filename) syntax are included in the serialized output

Example

Before this fix:

# Cell content: "Here is a header" (attachment link removed)
# Metadata still contains: { attachments: { "screenshot.png": "base64data..." } }
# Serialized .ipynb still includes the unused attachment data

After this fix:

# Cell content: "Here is a header" (attachment link removed)  
# Metadata contains: { attachments: { "screenshot.png": "base64data..." } }
# Serialized .ipynb excludes unused attachment data ✅

Changes

• serializers.ts: Added attachment filtering logic to createMarkdownCellFromNotebookCell() and createRawCellFromNotebookCell()
• serializers.test.ts: Updated existing tests and added new test cases for unused/partially used attachments

Testing

• Verified the exact issue scenario: attachment data is properly removed when references are deleted
• Tested edge cases: multiple attachments, partial deletion, complex filenames with spaces
• Ensured backward compatibility with existing attachment behavior

Fixes #200290.