Blank response if we fail to initialize about ngx_wasm_module

[thibaultcha]

I have rebased my proxy-wasm refactor branch atop main and now am blocked by the re-introduction of these test cases. Instead of producing an HTTP 500 at runtime, I think a better behavior is to exit the process altogether and prevent nginx from starting. Now, since initialization happens in init_worker, it cannot be tested since Test::Nginx cannot detect --- must_die when worker initialization failed (which is the reason why we don't check rc in (void) ngx_wavm_load() in init_process, which as recently attempted to change in #143). Today I'm looking at alternatives. I thought I had already solved this issue in the past by initializing the VM twice: once in the master process initialization to verify all modules are valid (and exit there if needed), and then a second time in init_process which would not fail if it was invoked (since master would have already exited). Maybe it was the V8 work that removed this behavior? I am less and less convinced that V8 introduced anything valuable to the project (especially since it isn't particularly more performant than other runtimes).

[hishamhm]

Yes, initializing the V8 engine twice doesn't work, it needs to be a global singleton.

[thibaultcha]

Fixed in #150 - we now prevent worker processes from starting when modules cannot be properly loaded.