MCP tool calls can move to the background as soon as a new message arrives

What

• MCP (Model Context Protocol) tool calls that run long now move to the background as soon as a new message needs to reach the session, instead of waiting for the existing auto-background timeout. The notification shown to the user explains the call was backgrounded so a waiting message could get through, and that the tool call itself was not interrupted and keeps running.
• A new foregroundCalls tracker, keyed by session, is registered when an MCP tool call starts (unless it's an inner/nested call), giving the auto-backgrounding logic visibility into how many calls are currently in the foreground for that session.
• Telemetry for auto-backgrounded MCP calls now records a trigger value distinguishing deliver_message from timeout.

Why This keeps a session responsive to new messages even while an MCP tool is still working, instead of forcing a wait for a fixed timeout before the session can move on.