📊 Telemetry: Profile `vscode.lm` proxy latency vs. API routing latency

[suhaibbinyounis]

Context

We recently shipped ExtensionHostProfiler.ts and PerfMetrics.ts to give us incredible native .cpuprofile metrics inside the node context.

The Missing Gap

Currently, when a high-latency spike occurs dynamically, it's hard to tell if the latency spawned from our stream restructuring loops or if the internal vscode.lm connection lagged talking to GitHub servers.

Proposed Solution

Add specific instrumentation spans covering the lmModel.sendRequest triggers and getCachedChatModels() API barriers. Output these distinctly into the dashboard telemetry counters so users can instantly know if VS Code itself is slow or if the API Gateway needs more optimization.