Add timeout option

[vandemjh]

Resolves timeout issues with older machines (especially those running on CPU) caused by a 300 second timeout constant.

• Removes whatwg-fetch import in favor of undici
• Adds a timeout prop to the request options, passing those to a new Agent

Resolves #72

	Bundle size (kB)
Before	70.2
After	71.2

Tests

On a crummy computer (or resource limited container) run:

require('ollama')
  .default.chat({
    model: 'gemma3:27b',
    messages: [{ role: 'user', content: 'Why is the sky blue?' }],
  })
  .then((i) => console.log(i));

Resulting in:

{
  "model": "gemma3:27b",
  "created_at": "2025-05-19T18:23:45.15035873Z",
  "message": {
    "role": "assistant",
    "content": "The sky is blue because of a phenomenon called **Rayleigh scattering**. Here's a breakdown..."
  },
  "done_reason": "stop",
  "done": true,
  "total_duration": 470459553167,
  "load_duration": 6352338645,
  "prompt_eval_count": 15,
  "prompt_eval_duration": 4274237804,
  "eval_count": 500,
  "eval_duration": 459831723963
}

7 minutes, nice!

[BruceMacD]

Hi @vandemjh, thanks for the contribution. One challenge we have here is using a default fetch library that is supported across different JS runtimes (versus only node). Seems like we do need a way to configure this, but it looks like using undici may cause issues for people using ollama-js outside a node environment. I'll test this out to confirm if that is the case.