Optimize document parsing for Banzai ReferenceCache

Problem

Banzai processing relies on a chain of filters. Each filter receives the output of the previous filter, then processes and returns the modified content.

We have multiple ReferenceFilters that include a ReferenceCache. It generates a cached collection of parent references by parsing the markdown document. The parsing process can be time and memory-consuming for large documents.

The biggest problem is that we repeatedly parse the document for each ReferenceFilter. That multiplies the overhead.

Proposal

• Avoid unnecessary document parsing calls
• Parse the document once and allow reference filters to access it