fix(errors): empty prompt logic

[rmay1er]

Bug description

agent.Generate returned an error when prompt was empty, even if user input
was already provided via the messages slice. This caused valid usage patterns
to fail unexpectedly.

---

Steps to reproduce (before)

		// Call the AI client to generate a response using the conversation messages.
		// Convert memorybox messages into the format expected by the AI client.
		resp, err := agent.Generate(ctx, fantasy.AgentCall{
			// Prompt:   input, // Was Error because empty.
			Messages: convert.ToFantasy(userMsgs),
		})
		if err != nil {
			// Print AI generation errors and stop.
			fmt.Printf("Generation error: %v\n", err)
			break
		}

Result:

error: prompt can't be empty

⸻

Expected behavior

If user input is already present in messages, an additional prompt string
should not be required.

⸻

Fix

•	Updated validation logic to return an error only when both prompt is empty and no messages are provided
•	Preserved existing system prompt resolution logic:
•	system message from messages is used when present
•	explicit system argument still overrides it

⸻

Result (after)

The same call now produces a valid prompt without returning an error.

⸻

Notes
• No breaking changes
• No public API changes
• CI and examples pass

---

• I have read CONTRIBUTING.md.
• I have created a discussion that was approved by a maintainer (for new features).

[kujtimiihoxha]

We should probably iterate over all messages and collect system role messages, + all message content instead of just the first message.

[kujtimiihoxha]

if prompt is empty we should guard that the last message in the message list is a user message, or a tool response.

I think its not expected that we send a message with the last message being an assistant message.

[kujtimiihoxha]

This seems to be also related to #59