Handle trigger breakpoints across process threads

[namtran1812]

Summary

Fix trigger breakpoints so they can fire from worker threads when magic-trace is attached to a multithreaded process.

Fixes #274.

Problem

The existing trigger implementation creates a hardware breakpoint only for the head PID/TID. Since perf_event_open breakpoints are task-specific, a trigger such as magic_trace_stop_indicator() can work from the main thread but be missed when called from a worker thread.

Implementation

• Enumerate process TIDs from /proc/<pid>/task
• Introduce a Trigger_breakpoint abstraction for breakpoint lifecycle and Async FD monitoring
• Maintain one trigger breakpoint per discovered TID
• Rescan for threads created after attachment
• Handle races where threads exit during breakpoint setup
• Preserve single-snapshot behavior across thread breakpoints
• Add a multithreaded regression target based on stop indicator does not work in multi-threaded application #274

Validation

• dune build @fmt passes locally
• git diff --check passes locally
• multithreaded C++ regression target builds successfully

End-to-end perf breakpoint behavior requires Linux validation, so this is being opened as a draft.

[namtran1812]

Draft for validation: DCO is passing, but the Linux build workflow is currently blocked before any jobs are created. I’ll keep this in draft until CI and end-to-end multithreaded perf validation are complete.