C++: Fix IR generation when ConditionDeclExpr does not have an immediate VariableAccess

[MathiasVP]

The QLDoc for ConditionDeclExpr:

/**
 * Gets the compiler-generated variable access that conceptually occurs after
 * the initialization of the declared variable.
 */
VariableAccess getVariableAccess() { result = this.getChild(0) }

made it seem like the first child was always a compiler-generated variable access of the declared variable. However, the test added in the first commit (identified in #15506) shows that this is not always the case.

This PR adds another predicate to ConditionDeclExpr for getting the 0'th child without casting it to a VariableAccess. We then use this new predicate in IR generation instead of getVariableAccess, and this fixes the bug exposed by the first commit.

Commit-by-commit review encouraged.

[geoffw0]

Looks great, fixes a bunch of consistency failures.

[MathiasVP]

Looks great, fixes a bunch of consistency failures.

Indeed! There are also some new results. Some of them are TPs and some of them are FPs. The FPs in particular are the three cpp/missing-check-scanf results on cmake-windows are FPs caused by guard conditions not being aware of switch statements.

I'll create an issue for this. I don't think this should block the PR.