Stop attach sessions when perf exits

[jeet-dekivadia]

Summary

• cache the perf recorder completion deferred so monitoring and final cleanup share a single waitpid
• wake attach sessions when their recorder exits after the tracee terminates
• tolerate ESRCH when snapshot or shutdown signals race an already-exited recorder

Why

In attach mode, killing the traced process makes perf record exit, but magic-trace only waits for Ctrl-C or a breakpoint snapshot. That leaves the command hanging after the tracee is gone. Sharing the recorder completion result also keeps shutdown from attempting a second reap.

Fixes #312

Validation

• git diff --check
• reviewed the attach wakeup and cached waitpid flow statically
• verified Perf_tool_backend.Recording is the only Backend_intf.S.Recording implementation and that monitoring plus final cleanup share the same cached waitpid deferred

[jeet-dekivadia]

Small follow-up for review context. This remains scoped to attach-session shutdown: Perf_tool_backend.Recording is the only recording implementation, and the patch caches its completion deferred so monitoring and cleanup observe the same single waitpid result. That lets a terminated tracee wake attach mode without changing Ctrl-C or breakpoint behavior, while the ESRCH handling only covers signal races after recorder exit.