Add a system prompt and stop_reason handling to the clients #274

Problem

None of the clients set a system prompt, and none of them check stop_reason. When the reply hits max_tokens or the model refuses, the user sees an empty or cut-off reply with no explanation. A quickstart should show the few lines that make this visible.

Tasks

• Add one short system prompt to all five clients, for example "You are a weather assistant. Use the tools to answer."
• After each API call, check stop_reason:
  • If it's max_tokens, print a one-line notice.
  • If it's refusal, print the refusal instead of an empty string.
• Keep the change to a few lines. Don't add streaming or typed error handling.

Acceptance criteria

• All five clients send the same system prompt.
• A reply cut off by max_tokens or refused by the model prints a notice instead of empty output.