feat: add tools to the agent briefing

[pkarw]

Related to #68, #63

Add kind of a agent briefing - where agents present the tools they have assigned to the supervisor prior to selecting the right agent for the task. I mean in the simplest way would be to modify the prompt in selectAgent: https://github.com/callstackincubator/ai-agent-framework/blob/bf188ffffbd8fdf1eaff8a7fb488a89837b1206a/packages/framework/src/supervisor/selectAgent.ts#39

[pkarw]

To implement this feature we could potentially modify selectAgent:

      {
        role: 'user',
        content: s`
          Here is the task:
          <task>${agentRequest.map((request) => request.content).join(',')}</task>

          Here are the available agents:
          <agents>
            ${agents.map(
              (agent, index) =>
                `<agent index="${index}">
                  <purpose>${agent.role} - ${agent.description}</purpose>
                  <tools>
                    ${
                      agent.tools
                        ? Object.entries(agent.tools)
                            .map(
                              ([name, tool]) => `<tool name="${name}">${tool.description}</tool>`
                            )
                            .join('')
                        : ''
                    }
                 </agent>`
            )}
          </agents>

          Select the most suitable agent for this task.
        `,
      },
      ```

[grabbou]

An alternative approach that I would like to experiment with, is for supervisor to be able to do a roundtrip with its agents to select the task, where agents can answer with confidence (e.g. score) how good they are to accomplish this task. Post-refactor, I would see that as a tool, which can be replaced to satisfy your own requirements. Supervisor would then, talk to all the agents and be able to get an answer, who's best to deal with it (this could also take into account "business" in the future, when we allow parallelism).

[grabbou]

This could also in the future take past self assessments into consideration, teaching supervisor that selected agent isn't as good as it thinks (confidence)

[pkarw]

This is great idea related to #72 - kind of reinforcement training