Leak a little less memory with dynamic allocations in Lua

[Mm2PL]

This PR ensures that any potentially large objects will get deallocated on error instead of getting forgotten and leaked.

This would have leaked memory:

function cmd_leak()
    local LARGE = string.rep("12345", 100)
    for i = 1, 1000000 do
        pcall(c2.register_command, LARGE) -- this errors and the error is thrown away
    end
end
c2.register_command("/leak", cmd_leak)

Now it doesn't leak this memory.

[github-actions]

clang-tidy made some suggestions

[github-actions]

warning: initializer for member 'errorReason' is redundant [readability-redundant-member-init]

Suggested change

	std::vector<QString> errorReason = {};
	std::vector<QString> errorReason;

[github-actions]

clang-tidy made some suggestions

[github-actions]

clang-tidy made some suggestions

[github-actions]

warning: function 'drop' has inline specifier but is implicitly inlined [readability-redundant-inline-specifier]

Suggested change

	inline void drop(T * /*var*/) = delete;
	void drop(T * /*var*/) = delete;

[Nerixyz]

You're invoking undefined behavior in many cases by using luaL_error which uses longjmp under the hood. On Windows, your approach can (and currently does; at least in the debug configuration) result in a double-free, as longjmp uses the same stack-unwinding as exceptions.

Speaking of exceptions... why not use them? At that point, it's probably best to use some bindings library (most (if not all) support incremental adoption). sol2 has a built-in trampoline. LuaBridge3 doesn't seem to have anything like it. There are probably others with support for that. I tried to convert the channel-api to use sol2 in https://github.com/Nerixyz/chatterino2/tree/experiment/sol3 (Nerixyz@45b7c0a) which has, despite adding adaptions for Qt types, more deletions than additions.

[Mm2PL]

Actually lua has an exception backend which could be used if we compiled it as c++ instead of c. This would then call destructors i think?

[Mm2PL]

I took a look at the sol this. And not having to do the SharedUserData is appealing but I'm not sure that it's appealing enough to rip everything out and convert it to sol.

[Mm2PL]

Closing in favor of moving everything to sol.