__thread var not respecting initialization when compiled with -O3

Please include the following in your bug report:

**Version of emscripten/emsdk:**
emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 3.0.0 (3fd52e107187b8a169bb04a02b9f982c8a075205)

**Failing command line in full:**
Compilation succeeds, but __thread var is not initialized

**Full link command and output with `-v` appended:**
I'm compiling several .o files before the final executable. Depending on how one of those .o files is compiled, the executable will or won't crash.

.o compilation that fails:
```
emcc -fPIC -O3 -fno-unsafe-math-optimizations -flto [... other -W and -I options ...] -c code_with_thread_volatile_var.c -o code_with_thread_volatile_var.o
```

.o compilation that succeeds:
```
emcc -fPIC -O0 -fno-unsafe-math-optimizations -flto [... other -W and -I options ...] -c code_with_thread_volatile_var.c -o code_with_thread_volatile_var.o
```

exe compilation (no change between success vs fail cases):
```
emcc -s EXPORTED_FUNCTIONS="['_free','_malloc' ]" -s EXPORTED_RUNTIME_METHODS="['setValue']" -s RESERVED_FUNCTION_POINTERS=4 -s ALLOW_MEMORY_GROWTH=1 -fPIC -O3 -DNDEBUG=1 -fno-unsafe-math-optimizations -mavx2 -flto -W... -I... -o /path/to/out.em.js [... various .o files including the one in question ...] -s FORCE_FILESYSTEM=1 -lidbfs.js
```

The following code change fixes this issue. Debugging the issue, it appears that without the below change, `v2` is not initialized to zero:

```
#if defined(__EMSCRIPTEN__) // added this to fix the issue
int64_t v2 = 0;
#else
__thread volatile int64_t v2 = 0; /* the issue seems to exist with or without the `volatile` keyword */
#endif
```

Admittedly there is some inconsistency in the fact that the code presumes a threaded environment where the exe compilation command does not include pthread or related options, but shouldn't the code still initialize properly nonetheless?