maxThinkingTokens thinking process is not visible in Agent SDK

When using maxThinkingTokens option with the Claude Agent SDK, the tags and their content are not exposed to developers, unlike when using the Messages API directly.

## Expected Behavior

When maxThinkingTokens is set, developers should be able to access the thinking process for each tool invocation, similar to how it works with the Messages API:

```json
{
role: "assistant",
content: [
{
type: "thinking",
thinking: "Let me analyze which tool to use..."
},
{
type: "tool_use",
// ...
}
]
}
```

## Current Behavior

The Agent SDK hides the thinking process entirely. Even when maxThinkingTokens is configured, only tool execution results are visible to developers.