Can you be more specific as to which type of tracing you are referring to? It's a highly overloaded term.
There's Go tracing (tracing the gVisor Go runtime), then there's syscall tracing (which can either mean "tracing the sandboxed application's system calls" or "tracing `runsc`'s own host system calls"), and then there's "I want to be able to trace this program that's running in gVisor from within gVisor" for any definition of "tracing". As I understand it, the "debug tracing" idea in the issue you linked to is about adding hypothetical seccheck events to the runsc code for debugging runsc, but these haven't been actually added.
Perhaps a more productive question: What is the use-case for the tracing operation you want to do? Is this about gVisor performance, or performance of an application within gVisor? Is it about one specific performance aspect (I/O speed/network speed/etc.) or about finding where a bottleneck is?