ai: return reasoning tokens in the streaming mode

[XiaochenCui]

fix #1323

Note that this PR is just a POC and should not merge before the ai module refactor is finished.

API request:

await puter.ai.chat('Tell me something I  might not know about chemistry.', {model: 'deepseek-reasoner', stream: true });

Here is the raw http response:

{"type":"reasoning","reasoning":""}
{"type":"reasoning","reasoning":"Okay"}
{"type":"reasoning","reasoning":","}
{"type":"reasoning","reasoning":" the"}
{"type":"reasoning","reasoning":" user"}
{"type":"reasoning","reasoning":" asked"}
...
{"type":"text","text":"Here"}
{"type":"text","text":"'s"}
{"type":"text","text":" a"}
{"type":"text","text":" fascinating"}
{"type":"text","text":" piece"}
{"type":"text","text":" of"}
{"type":"text","text":" chemistry"}

Code used in puter playground:

const reasoner_resp = await puter.ai.chat('Tell me something I  might not know about chemistry.', {model: 'deepseek-reasoner', stream: true });

for await (const part of reasoner_resp) {
  if (part?.text) {
    puter.print('<b>[Text]</b> ' + part.text.replaceAll('\n', '<br>'));
  }

  if (part?.reasoning) {
    puter.print('<b>[Reasoning]</b> ' + part.reasoning.replaceAll('\n', '<br>'));
  }
}