Why do you need to separately declare and then define every piece of your API? Because that's how C worked, and C did it that way because of the limitations of compilers in 1977. It's totally unnecessary in a modern language, and it makes your code way less clean. But that's how C++ works.
You don't have to. You can define all your code inside of the class, like in Java. It's up to the compiler/linker to store that mess somewhere. But for non-template code, it's more efficient (speed and memory) to have this separated, for both compiling and linking.
Why are templates designed in a way that makes you put the entire implementation in the header file? That was totally unnecessary, and it leads to clunky code. But that's how C++ works.
Because the compiling of the template is totally depending of the template arguments. Last time I've looked into Java, there were no templates (or even variadic templates), so this is really not comparable.
How come if a parent class doesn't mark its destructor as virtual, all subclasses will (silently) fail to get cleaned up correctly? This is just bad design. It's probably caused countless bugs over the years.
Only if the subclasses have virtual functions, then you should also make the destructor virtual. There are good reasons to have non-virtual destructors in parent classes. Why should I pay for a feature, if I don't need it? Here static code analysis helps.
The language is full of inconsistencies because no one ever bothered to fix them. Why is "this" a pointer instead of a reference?
Is this really an inconsitency? For me, pointers are no problem, they are an additional grade of freedom for expressing what I want. Yes, references are internally only pointers with the same speed and they normally don't have the value 0 (as long as you did not mess them up), but last is also true for 'this'. So it really doesn't matter.
Why does exception.what() return a char* instead of a string&?
Good question, but nearly irrelevant in well designed programs, that try to avoid exceptions during normal program flow. However, if you want to concatenate constant strings at compile time to get better error messages, there are tricks with variadic templates. With gcc, this even works with __PRETTY_FUNCTION__.
There are tons of minor points like this that could easily have been better if someone had bothered to think about consistency. But no one did. You can avoid the worst parts of C++, but what remains is still a poor substitute for a well designed language.
Perhaps. This language is for professionals. It is huge and there are pitfalls. But it gives you the neccessary freedom to express *exactly* what you want. C had been designed as a shortcut for Assembler and this is still true for C++; there is no virtual machine in between. If you write business applications, well, then use something else, but I do embedded development in the automotive sector and here C++ is exactly what I want.
Always look over your shoulder because everyone is watching and plotting against you.