CAPCOM-TD-OSS/REDox ? reverse-engineered prompt

Build me a .NET library that can read, edit, and write structured data really fast, with one shared model for JSON, JSON5, CBOR, MessagePack, TOML, XML, HTML, CSV, INI, and a custom DOX format.

I want it to feel like a mix of a fast document parser and an editable tree, so I can parse a file into a compact in memory representation, change values in place, then serialize it back out without rebuilding everything. It should support normal object serialization and deserialization too, plus easy conversion between formats. If it makes sense, include a small demo showing how to load JSON, change a few fields and array items, and save it again.

Please keep it focused on performance and low allocations, and make the API clean enough that someone could use it without learning a lot of internals. If you need to, look up the current .NET docs online before you finish.