fix: handle VS Code LM API token count of 4 by recalculating with string content

[roomote]

Fixes #6290

Summary

This PR addresses the issue where VS Code Language Model API was not correctly calculating the context size when using the "copilot - gpt-4.1" model. The problem was that the API would return a token count of 4 for LanguageModelChatMessage objects, which appears to be a placeholder value rather than the actual token count.

Changes

1. Modified internalCountTokens method: Added special handling when the token count equals 4 for LanguageModelChatMessage objects

   • Extracts the text content from the message
   • Recalculates the token count using the extracted string

2. Updated type checking: Changed from instanceof vscode.LanguageModelChatMessage to a duck-typing approach to work with both real and mocked objects

3. Added comprehensive test coverage:

   • Tests for normal token counting behavior
   • Tests for the special case when token count is 4
   • Tests for various edge cases and error scenarios

Implementation Details

As requested by @NaccOll, the solution:

• Still uses the internalCountTokens method
• Only when the text type is LanguageModelChatMessage AND tokenCount equals 4, converts the LanguageModelChatMessage to a string and recalculates via this.client.countTokens

Testing

• All existing tests pass
• Added new test cases specifically for this behavior
• Ran full test suite for the api/providers directory

This ensures that Roo Code continues to work correctly with VS Code's Language Model API and properly calculates context sizes.

---

Important

Fixes token count calculation in internalCountTokens for LanguageModelChatMessage when count is 4 by recalculating with string content.

• Behavior:
  • Fixes token count calculation in internalCountTokens in vscode-lm.ts for LanguageModelChatMessage when count is 4 by recalculating with string content.
  • Updates type checking from instanceof to duck-typing for compatibility with real and mocked objects.
• Testing:
  • Adds tests in vscode-lm.spec.ts for normal token counting, special case when token count is 4, and various edge cases.
• Misc:
  • Ensures all existing tests pass and runs full test suite for api/providers directory.

^{This description was created by for ce29a53. You can customize this summary. It will automatically update as commits are pushed.}

[daniel-lxs]

Closing since these are duplicates of #6424