To distinguish the SIGSEGVs that represent crashes from ones representing faults you care about, you look at the fault address. It works perfectly well: every high performance Java or C# runtime on Linux (e.g., HotSpot, ART) does it. You can trust the program, because SIGSEGV delivery isn't magic.

As I detailed in the doc and on libc-alpha, you really do need some kind of synchronous exception mechanism to match how real hardware behaves, and it would behoove libc authors to make this mechanism not suck instead of pretending that synchronous faults would just go away.