fix(js/genkit): correctly handle function prompt output schema

[pavelgj]

Fixes #1390

import { genkit, z } from 'genkit';
import { googleAI, gemini15Flash } from '@genkit-ai/googleai'

const ai = genkit({
  plugins: [googleAI()],
  model: gemini15Flash,
});

const helloPrompt = ai.definePrompt(
  {
    name: 'helloPrompt',
    input: z.object({
      name: z.string(),
    }),
    output: {
      schema: z.object({
        message: z.string()
      })
    }
  }, async (input) => ({
    messages: [{
      role: 'user',
      content: [{text: `say hello to ${input.name}`}]
    }]
  })
);


export const helloFlow = ai.defineFlow('hello', async () => {
  const { text } = await helloPrompt({name: "Pavel"});
  return text;
});

Checklist (if applicable):

• Tested (manually, unit tested, etc.)

[MichaelDoyle]

Thanks! This looks good.

As an aside, since functional prompts return a full generate request, technically the schema and other options can also be provided there. I wasn't sure how to reconcile that. :)

[pavelgj]

Thanks! This looks good.

As an aside, since functional prompts return a full generate request, technically the schema and other options can also be provided there. I wasn't sure how to reconcile that. :)

Good callout. I think most/all of these options are handled in wrapPromptActionInExecutablePrompt. I'm hunting down the ones I missed (or mishandled) one by one.