ChatDeepSeek translates `JSONDecodeError` on the sync path only — `ainvoke` / `astream` leak the raw "Expecting value"

langchain-deepseek 0.1.3 through 1.1.0 (latest, 2026-06-03) wraps only the sync _generate/_stream (langchain-ai/langchain#29758), so on a truncated or empty DeepSeek body invoke/stream say "DeepSeek API returned an invalid response. Please check the API status and try again." while ainvoke, astream, abatch and astream_events raise the bare JSONDecodeError("Expecting value"). No release fixes it (PR #40282 closed unmerged).

Workaround — subclass once and use it wherever you construct the model:

from json import JSONDecodeError
from langchain_deepseek import ChatDeepSeek

FRIENDLY = "DeepSeek API returned an invalid response. Please check the API status and try again."

class ChatDeepSeekAsyncSafe(ChatDeepSeek):
    async def _agenerate(self, *args, **kwargs):
        try:
            return await super()._agenerate(*args, **kwargs)
        except JSONDecodeError as e:
            raise JSONDecodeError(FRIENDLY, e.doc, e.pos) from e

    async def _astream(self, *args, **kwargs):
        try:
            async for chunk in super()._astream(*args, **kwargs):
                yield chunk
        except JSONDecodeError as e:
            raise JSONDecodeError(FRIENDLY, e.doc, e.pos) from e

All four async entry points go through these two methods. The exception type is unchanged, so existing except clauses keep working. Remove the subclass once a langchain-deepseek release translates the async path (the check below turns green then).

Reproduced 2026-09-11 in a fresh venv (Python 3.14.5, langchain-deepseek 1.1.0, langchain-openai 1.6.2, langchain-core 1.6.2, openai 3.13.0) with the issue's stub client — no API key, no network: stock class leaks on ainvoke and astream (2 leaks), the subclass reports the DeepSeek message on all four paths (0 leaks). Full record and repro script: docs/fixes/langchain-deepseek-jsondecodeerror-async-leak.md.