scxtop: Add perf event stats to TUI and MCP server

[hodgesds]

Overview

Add performance counter statistics (perf stat) integration for scxtop, including TUI visualization, MCP server API, and critical bug fixes of per-process monitoring.

TUI

• Real-time performance counter visualization with tabular and graphical views
• Derived metrics: IPC, cache/branch miss rates, stall percentages
• Multiple aggregation levels: system-wide, per-CPU, per-LLC, per-NUMA node
• Historical graphs using sparklines for trend analysis
• Integration with existing TUI navigation and key bindings

MCP Server API

• start_perf_stat: Start performance counter collection (system-wide or per-process)
• stop_perf_stat: Stop collection and finalize statistics
• get_perf_stat_results: Retrieve statistics with derived metrics
• Support for aggregation by system, CPU, LLC, NUMA node, and process
• Historical data collection for trend analysis
• Thread-safe SharedPerfStatCollector for concurrent access

Per-Process Monitoring Issues

• CPU filtering bug: Process events only monitored CPU 0 → Now monitors all CPUs with cpu=-1
• Inherit flag inverted: Child threads not tracked → Now correctly enabled for per-process monitoring

API Improvements

• Refactored PerfEvent to use i32 for CPU parameter (matches kernel API)
• Removed usize::MAX hack, replaced with explicit -1 for all-CPU monitoring
• Added comprehensive documentation with parameter descriptions
• More tests