Still needs the unit tests to be adjusted to test both implementations as well as making it properly selectable in the standalone binary.

This can possibly be introduced further by adding buffering to FileStream.

We could also add a MemStream version which works on a memory buffer allowing for the use case of just a passing a std::istringstream (i.e. string) and deprecating/removing the old API with std::istream.

Somehow using fgetc() seems to already interpret the characters causing 0x00 to be omitted when read even though the file is open in "binary" mode. Using fread() (which is supposed to be faster) fixes this but it is much slower.

Turns out it's just a bug in my code with unget() and multi-byte characters.

I backed out the test changes and will merge them into #261 so they don't get lost.

Testing with -q -Ilib/ -D__GNUC__ lib/valueflow.cpp:

current -> std::istream input -> FILE* input
Clang 14 263,400,261 -> 255,772,481 -> 241,915,419
GCC 12 266,098,949 -> 261,699,461 -> 247,682,262

Testing with -q -Ilib/ -D__GNUC__ lib/tokenize.cpp:

current -> std::istream input -> FILE* input
Clang 14 300,502,647 -> 294,632,802 -> 275,871,698
GCC 12 304,202,739 -> 301,878,425 -> 282,870,756

reducing the Ir count by using C I/O sounds good to me.

I put this into Cppcheck with the current and added TokenList construction and all tests passed on Windows and Linux. So this is finally ready for review.