A2A _serialize_value() corrupts JSON-native metadata types to strings

[YizukiAme]

Bug Description

_serialize_value() in from_adk_event.py falls through to str() for JSON-native Python types (dict, list, bool, int, float), corrupting structured metadata values into their string representations during A2A event conversion.

Steps to Reproduce

from google.adk.a2a.converters.from_adk_event import _serialize_value

metadata = {"retry": True, "max_attempts": 3, "tags": ["alpha", "beta"]}
result = _serialize_value(metadata)
print(type(result), result)
# Actual: <class 'str'> "{'retry': True, 'max_attempts': 3, 'tags': ['alpha', 'beta']}"
# Expected: <class 'dict'> {'retry': True, 'max_attempts': 3, 'tags': ['alpha', 'beta']}

Expected Behavior

JSON-native types (dict, list, str, bool, int, float) should be preserved as-is since they are already valid JSON values, not converted to strings.

Actual Behavior

All non-Pydantic values that aren't None are passed through str(), which corrupts structured data into Python repr strings.

[enjoykumawat]

I'd like to work on this. The fix is to pass through JSON-native types (dict, list, int, float, bool) as-is in _serialize_value() instead of stringifying them. Will submit a PR shortly.

[sanketpatil06]

Hi @YizukiAme,
Thank you for reporting this issue. I have reproduced it on my end and can confirm the metadata serialization behavior you are seeing.
Additionally, thank you both, @YizukiAme and @enjoykumawat, for submitting pull requests. Our team is currently reviewing both PRs to determine the best approach for a fix.
We really appreciate your contributions.

[enjoykumawat]

Thanks for the update @sanketpatil06! Happy to hear the team is reviewing both approaches.

I'm happy to coordinate with @YizukiAme — if their implementation fits better with the codebase direction, feel free to close #5200. The goal is just to get the fix in.

If there's anything specific you'd like me to adjust on my PR to make it easier to evaluate, just let me know.