fix(mcp): Pass working directory to spawned qsv processes

[Copilot]

The MCP server was spawning qsv processes without setting their working directory, causing them to run in the Node.js process's cwd instead of the configured working directory. This resulted in secondary output files (e.g., *.stats.bivariate.csv) being written to unexpected locations.

Changes:

• SkillExecutor: Added workingDirectory property, passed as cwd to spawn(), with getter/setter and debug logging
• QsvMcpServer: Initialize executor with config.workingDir, sync working directory when qsv_set_working_dir is called
• executePipeline: Pass working directory from filesystem provider or config to executor to ensure pipeline steps run in correct directory

// Before: executor created without working directory
const executor = new SkillExecutor(config.qsvBinPath);

// After: executor uses configured working directory
const workingDir = filesystemProvider?.getWorkingDirectory() || config.workingDir;
const executor = new SkillExecutor(config.qsvBinPath, workingDir);

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.