AgentCoreMemorySessionManager does not support extended thinking

**Describe the bug**
Multi-turn interaction with an agent using a model with [extended thinking](https://docs.claude.com/en/docs/build-with-claude/extended-thinking) does not work

```
An error occurred (ValidationException) when calling the ConverseStream operation: The model returned the following errors: messages.11.content.0.type: Expected `thinking` or `redacted_thinking`, but found `text`. When `thinking` is enabled, a final `assistant` message must start with a thinking block. We recommend you include thinking blocks from previous turns.
```

**To Reproduce**
Steps to reproduce the behavior:
* Create a Strands Agent with AgentCoreMemorySessionManager and with Claude Sonnet 4.5 with extended thinking enabled
```
additional_request_fields={
"thinking": {"type": "enabled", "budget_tokens": 4096}
}
```
* Interact with the agent twice in the same session

**Expected behavior**
Multi-turn interaction works without an error

Start with AgentCoreMemorySessionManager and reproduce the two-turn interaction using Claude Sonnet 4.5 with extended thinking and the shown additional_request_fields. Trace how messages are persisted and restored between turns; done means the second interaction completes without the ConverseStream validation error.

Written by the indexing model from the issue text.