WISH: ndjson codec

[hiiamboris]

From time to time I happen to need to work with ndjson data. It is trivial enough to de/encode via json codec, but it seems to be common enough to have out of the box support.

[dockimbel]

I see that used in LLM servers API, so pretty widely used nowadays.

I guess that should be supported as a port scheme, inheriting from the HTTP scheme.

[greggirwin]

There are two sides to this, right? The port scheme part, and codec part. It also affects how data is structured for "record/line oriented" use, but that's a different discussion. What this does is make JSON more like CSV in terms of /lines being implied. This is all good. The next question is whether we want to think about a Red version of this as well. I have a very old single-line-mold R2 func I used a lot in one particular system.

Our JSON codec tracks newlines as it parses, so it can provide better error information. Structured newline-delimited data changes how we think about data. I don't know, with NDJSON, if large, complex records are common. Our CSV codec is very forgiving of errors like mismatched/unclosed quoted fields, so basically looks at errors at a higher level, matching field counts and such. For line-oriented formats, we may want to think about row/col offsets for parse error info.