aws: Bedrock LLM logs a WARNING "unknown chunk type" for every reasoning delta (seen with OpenAI GPT-6, gpt-oss, Claude thinking)

[kimnamu]

Bug Description

Thanks for the Bedrock plugin. While testing OpenAI GPT-6 on Bedrock through aws.LLM, everything worked apart from a warning that shows up on most turns.

ConverseStream sends reasoning as contentBlockDelta.delta.reasoningContent: redactedContent bytes for GPT-6, text/signature for gpt-oss and Claude extended thinking. LLMStream._parse_chunk only handles toolUse and text deltas (llm.py L360-L368). Every reasoning delta falls into the else branch and logs WARNING aws bedrock llm: unknown chunk type, with the raw chunk in lk.pii.chunk. The response text is still correct.

model, us-east-1, one 250-word answer	"unknown chunk type" warnings on main
us.openai.gpt-6-luna, default effort	1
us.openai.gpt-6-luna, reasoning.effort=high	2
us.openai.gpt-6-sol, default / high	0 / 1
openai.gpt-oss-120b-1:0	1
us.anthropic.claude-sonnet-4-6 with thinking enabled	3 (2 text, 1 signature)

GPT-6 Luna hits this at default settings: 6 of 6 short prompts at default effort logged it (Sol: 1 of 6).

Expected Behavior

Reasoning deltas are skipped quietly, the way the anthropic plugin skips thinking deltas (anthropic/llm.py L367-L388). The warning stays for delta types that really are unknown.

Reproduction Steps

cat > repro.py <<'PY'
import asyncio, logging
from livekit.agents.llm import ChatContext
from livekit.plugins import aws

logging.basicConfig(level=logging.WARNING)

async def main():
    ctx = ChatContext()
    ctx.add_message(role="user", content="Write about 250 words explaining how a hash map works.")
    await aws.LLM(model="us.openai.gpt-6-luna").chat(chat_ctx=ctx).collect()

asyncio.run(main())
PY
uv run python repro.py

WARNING:livekit.plugins.aws:aws bedrock llm: unknown chunk type

Operating System

macOS 26 (not OS-specific)

Models Used

AWS Bedrock us.openai.gpt-6-luna, us.openai.gpt-6-sol, openai.gpt-oss-120b-1:0, us.anthropic.claude-sonnet-4-6 via aws.LLM

Package Versions

livekit-agents==1.8.2
livekit-plugins-aws==1.8.2   # main at b4df92e
aiobotocore==3.8.0
botocore==1.43.46
python==3.13

Session/Room/Call IDs

N/A, reproduces with aws.LLM alone.

Proposed Solution

            elif "text" in delta:
                ...
            elif "reasoningContent" in delta:
                # Reasoning deltas (text/signature, or the redactedContent OpenAI GPT-6
                # streams) are not surfaced; skip them quietly, as the anthropic plugin
                # does for thinking deltas.
                pass
            else:
                logger.warning("aws bedrock llm: unknown chunk type", extra={"lk.pii.chunk": chunk})

I have this with a hermetic test and can open a PR.

Additional Context

Delta shape

ReasoningContentBlockDelta in the botocore bedrock-runtime model has text, redactedContent and signature. The chunk GPT-6 Luna logged:

{'contentBlockDelta': {'delta': {'reasoningContent': {'redactedContent': b'rsn_rru6BQHg...'}}, 'contentBlockIndex': 0}}

The count per answer varies from run to run; it always equals the number of reasoning deltas in the stream.

Surfacing reasoning as its own stream would be a feature, not this fix.