Fix race on fatal signals - #294

This PR proposes a fix for a race condition that appears when a process has multiple threads.
If we resume one thread and deliver a fatal signal (e.g., SIGSEGV), the kernel may start the thread-group exit before we resume the other threads. As a result, ptrace(PTRACE_CONT) on those threads fails because they have already begun exiting.

What this PR does

• Detects when a signal is truly fatal to the process.
• If a fatal signal is pending, only the affected thread is resumed.
• Other threads remain stopped, avoiding the race and preventing PTRACE_CONT failures.

Status

Draft for two reasons:

• The approach works in initial testing but needs broader validation.
• It builds on Add exec/execv/execve support #273, which is also a draft (we can cherry-pick if needed).
• I'm not totally convinced of some details.