fix(Format): Extract correct audio language from captions

[absidue]

Turns out the properties I was using before, are influenced by the language you request the page in and have nothing to do with the language of the audio. Instead we need to check the language of the auto-generated captions track, because that is generated in the language of the audio, yt-dlp does the same.

The video provided in #488 can be used to demonstrate the issue with the code below. It previously printed en but will correctly print ru, with this change. The video in question is in Russian, but because it has an English caption track and we requested the page in English, YouTube makes the English caption track the default.

import { Innertube } from './dist/src/platform/node.js';

const session = await Innertube.create({ lang: 'en' })
const info = await session.getBasicInfo('ID7VkY6cMRQ')

console.log(info.streaming_data.formats[0].language)

This fixes bugs in these pull requests #470 and #445.