Python: Fix Bedrock non-ASCII escaping in JSON content blocks

[kimnamu]

Thanks for Agent Framework and the Bedrock integration — it's a pleasure to build on.

Closes #6627

Problem

When the Bedrock Converse API returns a structured json content block, BedrockChatClient._parse_message_contents serializes it to text with json.dumps(json_value). Since json.dumps defaults to ensure_ascii=True, non-ASCII characters (CJK, emoji, accented Latin, etc.) are escaped to \uXXXX and reach the user garbled.

Cause

This single call is the outlier. The sibling OpenAI client (python/packages/openai/agent_framework_openai/_chat_client.py) and 16+ other call sites across the repo already serialize user-facing / span data with ensure_ascii=False. PR #3894 ("Python: Fix non-ascii chars in span attributes") fixed the same class of issue in observability — this is the matching fix the Bedrock client was missing.

Change

One line: add ensure_ascii=False to the json.dumps for the Bedrock json content block, plus a regression test.

Before / After

Item	Before	After
json block {"greeting": "你好世界"} → text	❌ {"greeting": "你好世界"}	✅ {"greeting": "你好世界"}
Emoji "🎉" in json block	❌ "🎉"	✅ "🎉"
Text content blocks (block.get("text"))	✅ unchanged	✅ unchanged
Public API / method signatures / arguments	✅ unchanged	✅ unchanged
Output remains valid JSON (round-trips via json.loads)	✅	✅

Test (red → green)

New test test_process_converse_response_preserves_non_ascii_in_json_block uses the existing key-free _StubBedrockRuntime pattern.

Against the unpatched source (bug present):

>       assert "你好世界" in text
E       assert '你好世界' in '{"greeting": "\\u4f60\\u597d\\u4e16\\u754c", "emoji": "\\ud83c\\udf89"}'
FAILED packages/bedrock/tests/test_bedrock_client.py::test_process_converse_response_preserves_non_ascii_in_json_block
======================== 1 failed, 2 warnings in 0.29s =========================

With the fix:

packages/bedrock/tests/test_bedrock_client.py .                          [100%]
======================== 1 passed, 2 warnings in 1.88s =========================

Full bedrock suite (no regressions) + lint/format clean:

$ uv run --package agent-framework-bedrock pytest packages/bedrock/tests/ -m "not integration" -q
..................................                                       [100%]
$ uv run ruff format --check packages/bedrock/...   # 2 files already formatted
$ uv run ruff check packages/bedrock/...            # All checks passed!

---

This contribution was prepared with the help of an AI agent (Claude Code); a human reviewed the change, rationale, and test results before submitting.

[Copilot]

Pull request overview

Fixes a Bedrock Converse response parsing edge case where structured json content blocks were being serialized with json.dumps(..., ensure_ascii=True) (default), causing non‑ASCII characters to be escaped into \uXXXX sequences in user-visible text.

Changes:

• Update Bedrock json content block serialization to use ensure_ascii=False so non‑ASCII characters are preserved.
• Add a regression test ensuring CJK text and emoji remain unescaped and the emitted text still round-trips via json.loads.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/packages/bedrock/agent_framework_bedrock/_chat_client.py	Preserves non‑ASCII characters when converting Bedrock json content blocks into text.
python/packages/bedrock/tests/test_bedrock_client.py	Adds a unit test verifying non‑ASCII preservation and preventing regression.