Ollama/Mistral Fixes and Improvements

[lazydog2]

I wasn't able to get the movies.run.ts example to work as is and had to change "template:" to "prompt:" here:
https://github.com/hrishioa/wishful-search/blob/master/src/ollama.ts#L24

I was also experiencing UND_ERR_HEADERS_TIMEOUT errors because the LLM was taking longer to respond than the fetch default of 30 seconds so I made the following changes:
Added the "undici" dependency to package.json and ran npm install:

"dependencies": {
    "@anthropic-ai/sdk": "^0.16.1",
    "@azure/openai": "^1.0.0-beta.9",
    "openai": "^4.24.7",
    "sql.js": "^1.8.0",
    "undici": "^6.19.0",
  },

Added the following to the beginning of ollama.ts:
import { fetch, Agent } from 'undici';
And replaced:

// Using fetch to initiate a POST request that will return an SSE stream
  const response = await fetch(`http://127.0.0.1:${port}/api/generate`, {
    method: 'POST',
    headers: headers,
    body: requestBody,
  });

With:

// Using fetch to initiate a POST request that will return an SSE stream
  const response = await fetch(`http://127.0.0.1:${port}/api/generate`, {
    method: 'POST',
    headers: headers,
    body: requestBody,
    dispatcher: new Agent({
      headersTimeout: 10 * 60e3,  // 10 minutes
      bodyTimeout: 10 * 60e3,     // 10 minutes
    }),
  });

I also found that when the response from the LLM contained a curly brace, the json parsing would fail so I replaced the following in ollama.ts:

      // Attempt to parse all complete JSON objects in the buffer
      while (true) {
        const openingBraceIndex = textBuffer.indexOf('{');
        const closingBraceIndex = textBuffer.indexOf('}');

        // Check if we have a complete object
        if (openingBraceIndex !== -1 && closingBraceIndex !== -1) {
          // Extract and parse the JSON object
          const jsonString = textBuffer.slice(
            openingBraceIndex,
            closingBraceIndex + 1,
          );

With (there may be better ways to achieve this):

      const regex = /{.*?"done":(?:true|false).*?}/g;
      const jsonStrings = textBuffer.match(regex);
      if (jsonStrings == null) {
        continue;
      } else if (jsonStrings.length == 0) {
        continue;
      }

      // Attempt to parse all complete JSON objects in the buffer
      for (let i = 0; i < jsonStrings.length; i++) {
        const currentJsonString = jsonStrings[i];
	if (currentJsonString == null) {
            break;
        }
        const closingBraceIndex = currentJsonString.length;

        // Check if we have a complete object
        if (closingBraceIndex !== -1) {
          const jsonString = currentJsonString;