Add chat performance benchmarking harness

Introduces an end-to-end chat performance benchmarking and memory leak detection framework, backed by a deterministic mock LLM server and a CI workflow for automated regression testing.

What's included

Perf regression runner (npm run perf:chat)

• Launches VS Code via Playwright Electron, opens the chat panel, sends messages with a mock LLM response, and measures timing, layout, rendering, and memory metrics
• Compares a test build against a baseline (defaults to latest stable release) using Welch's t-test for statistical significance
• Supports dev, production (--production-build), and release builds with mismatch detection
• Resumable runs (--resume) to accumulate samples for higher confidence
• Per-build settings overrides for A/B testing experimental features

Memory leak checker (npm run perf:chat-leak)

• Cycles through all registered scenarios in a single session, measuring heap and DOM node growth between iterations
• Uses linear regression on post-GC heap samples to detect sustained per-iteration leaks

Mock LLM server (mock-llm-server.js)

• Implements the full CAPI URL structure (/models, /models/session, /chat/completions, etc.) for deterministic, zero-latency responses
• Supports content streaming, tool calls, thinking blocks, and multi-turn conversations via a scenario-based architecture

CI workflow (chat-perf.yml)

• workflow_dispatch with configurable inputs (baseline/test commits, runs, threshold, settings overrides)
• Builds once, then fans out perf comparison across 4 matrix groups + a parallel leak check
• Merges per-group results into a unified Markdown summary posted to the GitHub step summary
• All user inputs passed via environment variables (not direct ${{ inputs.* }} interpolation) to prevent script injection

Scenarios (perf-scenarios.js)

• Content-only: text-only, large-codeblock, rapid-stream, mixed-markdown
• Tool-call: tool-read-file, tool-edit-file, tool-terminal
• Multi-turn: thinking-response, multi-turn-user, long-conversation

Other changes

• package.json: Added perf:chat and perf:chat-leak npm scripts
• SKILL.md: Agent skill documentation for running benchmarks