[Codestyle] Set clang-format RemoveSemicolon rule to true

[adamscott]

• Introduces the FORCE_SEMICOLON macro to force a semi-colon to be inserted (and bypass clang-format).
  • See [Codestyle] Set clang-format RemoveSemicolon rule to true #97934 (review) for an example.
• Sets clang-format Standard rule to c++20 (as there's currently a c++20 feature in the codebase that messes up with semi-colons otherwise)

This should fix the superfluous semi-colons issue that takes a lot of time for some reviewers to flag in PRs.

[AThousandShips]

See also:

• Style: Detect unnecessary semicolons via clang-tidy #97241

For a different approach (which doesn't apply to CI but allows us to fix this)

[adamscott]

Here's an example of the use of FORCE_SEMICOLON. Only use when the formatter is stubborn removing a necessary semi-colon.

[Repiteo]

Alternatively:

	SCOPE_EXIT { // clang-format off
		_recursion_flag_ = false;
	}; // clang-format on

[adamscott]

I don't know. Like, yeah it works, but what if a format error is introduced between them?
What I mean is that it's the only case in the whole codebase that it occurs.

[Repiteo]

If format errors in between are the concern, just move the // clang-format off down one line. Because the settings only apply to subsequent lines, this will turn clang-format off only for the line with the semicolon.

	SCOPE_EXIT {
		_recursion_flag_ = false; // clang-format off
	}; // clang-format on

While that's the option I'd prefer most, here's two more alternatives: after looking at the SCOPE_EXIT macro, it turns out this is the only place in the entire repo where it's used. I'd sooner say to either remove that macro entirely, or replace it with a version that wraps the content itself, both of which will fix clang-format removing the semicolon.

	gdmono::ScopeExit GD_UNIQUE_NAME(gd_scope_exit) = gdmono::ScopeExitAux() + [=]() -> void {
		_recursion_flag_ = false;
	};

#define SCOPE_EXIT(m_content) \
	auto GD_UNIQUE_NAME(gd_scope_exit) = gdmono::ScopeExitAux() + [=]() -> void { m_content };
//////////
	SCOPE_EXIT(_recursion_flag_ = false;)

---

I prefer these workarounds because I see FORCE_SEMICOLON better used for ensuring a trailing semicolon on macro. Here's an excerpt from one of my prospective PRs that wanted to tackle this on the clang-tidy scope:

// By appending to a macro, ensures that it'll require a trailing semicolon.
#ifndef FORCE_SEMICOLON
#define FORCE_SEMICOLON static_assert(true)
#endif

static_assert produces zero build-time bloat, being a functionally "free" addition (especially in this always-true context). What it does provide is a way to require a trailing semicolon in virtually any context, as static_assert requires a semicolon to follow its declaration.

[Repiteo]

I disagree with FORCE_SEMICOLON being added in this context (see #97934 (comment)), but I'm otherwise onboard with this! I'd recommend a .git-blame-ignore-revs commit as well, as there's no functional changes to speak of.