Make jobs recoverable and log handling scalable

[magetsu002]

This PR improves job recovery and log handling across Chronosphere.

What changed

• persists each subprocess PID, process-group ID, startup token, and start time in a private runtime sidecar;
• validates recovered process identity before sending signals, protecting against PID reuse;
• reconciles jobs marked as running when an engagement or MCP session starts;
• restores control of valid jobs that survived a restart and marks dead or mismatched jobs accurately;
• supports safe TERM/KILL cancellation for recovered jobs, not only jobs started in the current session;
• replaces whole-file log tailing with bounded seek-from-end reads;
• replaces whole-file grep with buffered, byte-bounded streaming and capped results;
• handles large lines and non-UTF-8 output without unbounded memory use;
• adds subprocess, process-group, restart recovery, large-log, result-cap, and truncation tests;
• expands doctor to report stale running jobs and orphaned runtime, status, SSH password, remote script, and log files;
• adds repair support that removes disposable artifacts and archives orphan logs safely.

No command-library expansion or TUI redesign is included.

Validated with formatting, Clippy, the full locked test suite, and a locked release build.