Log token budget and usage for Copilot tools

[minestarks]

This way we can keep an eye on how various use cases and instructions files affect the token budget for our tools.

While this number doesn't give us true visibility into the internals of Copilot, like the global token budget, it may at least act as a rough indicator of how much context we're eating up.

Some numbers I gathered to help paint a picture:

• We seem to get about ~43000 tokens in our budget for each tool call. Every message in the conversation history seems to reduce this budget, but not by much - by about 2
• When we include our Q# instructions file in the context, the token budget seems to decrease about ~500 tokens
• Claude talks more than GPT 4.1, thereby reducing our budget slightly 🙄
• A typical short output, like the circuit from a Bell state program) takes up ~200 tokens

[minestarks]

Thanks @billti for the review