Simplify API by decreasing a number of entry points

[foxtran]

I have removed most of TRACY_HAS_CALLSTACK in C/C++ public headers with implementing related constexpr (if possible) checks in functions' implementations.

I have also set TRACY_CALLSTACK to zero if it is not set. It is set to zero, stack will not be collected. I hope compiler will be able to determine that corresponding branches are always false and remove related code.

Note: only C/C++ headers are updated. I did not touch other APIs which can be also reduced in the same way.

Fixes #954

[wolfpld]

Can you show disassembly of the code that would use functions you have changed? I don't really want the if depth > 0 check to be present, if it's known at compile time it's not needed.

[foxtran]

I do not know how to show it in a good way...

Anyway, there is a relatively simple example: https://godbolt.org/z/5h7jz5dzz

I have inserted exit(depth) calls into public/client/TracyScoped.hpp in two branches. As you can see, it will generate an extremely simple code that will only pass 0 or 5 (I set these depths) into exit call with -O3 optimization level. With -DCMAKE_BUILD_TYPE=Debug, TRACY_CALLSTACK=0 will get a cost but starting from -O1 it will work with a simple asm :)