Revert back to iterating over lines

[sjmonson]

Summary

Partially reverts #663 to iterating over lines, but keeps the skipping of blank newlines.

Details

#663 switched the HTTP backend to iterating over byte strings. The problem is that is did not handle the case where a line was split over multiple iterations.

Test Plan

Run a benchmark with known errored request rate (preferably 0) and ensure that there are no failed requests due to orjson.JSONDecodeError: unexpected end of data.

---

• "I certify that all code in this PR is my own, except as noted below."

Use of AI

• Includes AI-assisted code completion
• Includes code generated by an AI application
• Includes AI-generated tests (NOTE: AI written tests should have a docstring that includes ## WRITTEN BY AI ##)

[Copilot]

Pull request overview

This PR adjusts the OpenAI HTTP streaming implementation to iterate over decoded text lines again (instead of splitting raw byte chunks), while still skipping blank newline separators, to avoid JSON parsing failures when a single SSE line is split across multiple network reads.

Changes:

• Switch _aiter_lines() from Response.aiter_bytes() + split(b"\n\n") to Response.aiter_lines()
• Preserve behavior of skipping blank/whitespace-only lines during streaming

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[jaredoconnell]

Looks fine. Are there any potential problems with skipping blank lines?

[dbutenhof]

Let's get this in and freeze v0.6.

[sjmonson]

Are there any potential problems with skipping blank lines?

There shouldn't be. Its mainly to account for the fact that every data portion ends with \n\n. So we were counting the extra newline as an iteration even though its just an end indicator for the data. According to the spec here the extra newline is just to separate multi-part event into blocks of related information. E.g.

data: first event
id: 1

data:second event
id

data:  third event
data: more data

It does not seem like any of the OpenAI endpoints use this feature though since all of the data is on one line.