Add max_tokens parameter support to runWithTools

[koogunmo]

Description:

This adds support for controlling the maximum number of tokens in AI responses by passing an optional max_tokens parameter to runWithTools.

Problem:
The default token limit of 256 causes response truncation for longer outputs. There's currently no way to configure this.

Solution:

• Add max_tokens parameter to runWithTools input options
• Pass max_tokens to both AI.run() calls (initial and final response)
• Create shared ModelName type (keyof AiModels) to replace non-existent BaseAiTextGenerationModels
• Update type references in runWithTools.ts and utils.ts

Usage:

  const response = await runWithTools(
    env.AI,
    '@cf/meta/llama-3.3-70b-instruct-fp8-fast',
    {
      messages,
      tools: [searchTool],
      max_tokens: 2048,  // Now supported
    }
  );

Testing:
Tested in production with personal application. Prevents truncation of longer AI responses while maintaining backward
compatibility (parameter is optional).