You can actually do pretty much anything you want, it's just the C library that uses a lot of global state and internal memory allocations, which messes things up. The core syscall API and any reentrant code you write yourself are not affected.
>The memory state that a program receiving a SIGSEGV in is often undefined/garbage
That may be true for arbitrary segfaults caused by bugs, but the JIT has 100% control over what instructions to emit, it is not restricted by ABIs or platform-specific issues, so there is no problem to use SEGV as a signaling mechanism.