feat: Add per-request metadata support for MCP operations

Use case:
We have a Customers MCP. In our organization, each user gets access only to certain customers. Our agents are in a session-managed multi-user environment, so we needed a way to send user details on a per-request basis (Headers are supported globally, but only set at initialization, leaving no way to send different headers with different requests). While it would be possible to send this information as tool call parameters for example, this really would bloat our tool schemas.

Implementation:
- Added RequestParams.mcp_metadata field for passing custom metadata
- Uses contextVar at augmented_llm level to store metadata across async tasks
- Extends agent method signatures to accept RequestParams (send, interactive, prompt, etc.)
- Enhanced MCPAgentClientSession with _meta parameter support for call_tool, read_resource, get_prompt
- Implements safe merging to preserve existing meta fields (e.g. progressToken)
- Metadata flows from RequestParams → Context Variable → MCPAggregator → MCPAgentClientSession → MCP Server

Usage:
```python
request_params = RequestParams(mcp_metadata={"user_id": "123", "auth0_id": "auth0|..."})
await agent.generate("Get customers", request_params)
# MCP server can access via: ctx.request_context.meta.user_id
```

Supports all agent methods and MCP operation types (tools, resources, prompts).

internal ticket: DEV-1888

add the new request_parameters to the send() overrides to fix this test