Bug: Include Preprocessing reformats include file's source

From examining `glslang/MachineIndependent/ShaderLang.cpp` it seems apparent that the entire include file is parsed (obviously so the preprocessor directives can be honoured), **but then the actual source produced from all the headers is de-parsed into a string of formatted tokens**

This is a problem for me as I outline in this shaderc issue:
https://github.com/google/shaderc/issues/1069

Long story short :
> I have a peculiar method of disabling preprocessor directives and checks for GL_ extension defines until after the #includes have been collected. I replace all preprocessor directives except for #version and #include with _this_is_a_hash and some programmatically generated guards against multiple inclusion.

This is working for me quite well, but the only place where it breaks are:

- `#defines` that have a () after them, such as `_this_is_a_hashdefine ONE_MORE_THAN_A (A+1)`
- preprocessor function macros such, ex. `_this_is_a_hashdefine FUNC(X) (X*X)`

The cause of all this is because your `PreprocessDeferred` re-glues together the headers from tokens and the `(` and `)` symbols are on some sort of a list of tokens that are explicitly stated not to need a space before them.

```cpp
// Output a space in between tokens, but not at the start of a line,
// and also not around special tokens. This helps with readability
// and consistency.
if (!isNewString && !isNewLine && lastToken != EndOfInput &&
(unNeededSpaceTokens.find((char)token) == std::string::npos) &&
(unNeededSpaceTokens.find((char)lastToken) == std::string::npos) &&
(noSpaceBeforeTokens.find((char)token) == std::string::npos)) {
outputBuffer += ' ';
}
```

Is there any way to fix this behaviour?
could I remove `(` from `noSpaceBeforeTokens` without affecting somethng else adversely?

Start in glslang/MachineIndependent/ShaderLang.cpp, focusing on PreprocessDeferred and the token-spacing logic involving noSpaceBeforeTokens. Reproduce the include cases with function-like macros and parenthesized macro bodies, then verify that preprocessing preserves the required separation without changing other formatted output.

Written by the indexing model from the issue text.