Chat process follow-ups: a mid-stream error raises, .events survives partial reads, bad show_process chokes first

[rejojer]

Review fixes for the chat-process feature (v0.2.13), re-applied on main. They sit on the review view #457 as d5110fb and 82b7558; chat() was reworked by #460 since, so the client.py hunks were merged by hand onto the current wording.

A mid-stream error chunk now raises. The managed endpoint reports a server-side failure as a final {"error": ...} chunk after the partial answer (it refunds the credits, then yields it). Neither SSE decoder in cloud_api looked at it, so chat(stream=True) and chat_completions(stream=True) in both modes ended as a short, apparently complete answer with no exception. One guard in each decoder raises PageIndexAPIError; the partial answer is still delivered first.

.events survives partial reads. ChatStream.events delegated with yield from, so a dropped handle (next(stream.events), for ... break) closed the shared run on GC and the rest of the run silently vanished. A plain loop leaves it alone.

Hidden call lines still label results. _weave recorded a call's arguments only past the tool_call visibility guard, so with call lines hidden the standalone [tool_result] lines never carried the arguments they promise.

A bad show_process chokes first. The value is validated before the stream check, so an invalid value is refused as such instead of being told to add stream=True and then refused again; the managed lane's duplicate choke goes with it.

Docs. stream: and the Returns block say show_process is on by default and show_process=False gives the bare answer; the show_process opener no longer says own-model only.

Verification: 475 passed; whole suite with openai-agents blocked 369 passed / 106 skipped; pyright client.py 0, package unchanged (236). The error-chunk test loops all three streaming surfaces and was red before the guard; removing either guard alone fails it.

https://claude.ai/code/session_01PYr9yG1FPQxKCA9m7ECQWY