Claude sessions pid matching

[codeaholicguy]

Summary

• Claude Code (newer versions) writes ~/.claude/sessions/.json for each running process, containing { pid, sessionId, cwd, startedAt }. This PR updates ClaudeCodeAdapter to use these files as the primary matching source instead of the existing CWD-encoding + session file birthtime heuristic.
• For each discovered Claude process, the adapter now attempts to read ~/.claude/sessions/.json directly. If found and valid, the sessionId is used to locate the JSONL file exactly.
• Processes without a PID file (older Claude Code versions, or files not yet written) fall back to the existing legacy matching algorithm unchanged.
• A stale-file guard (60 s tolerance between startedAt and proc.startTime) protects against PID reuse.

Why

The previous birthtime-based heuristic could produce incorrect matches when multiple Claude instances ran in the same working directory, or when session file creation time diverged from process start time. The new approach is exact and deterministic.