C++: Add PostUpdateNode for updates to structs with no chi instructions

[MathiasVP]

In #3118 we added basic IR field flow by attaching PostUpdateNodes to chi instructions. However, updating a field in a struct doesn't always generate a chi instruction. In particular, writing to a struct with a single field doesn't generate a chi instruction. This means we don't catch the following dataflow:

struct A {
  int i;
};
void foo() {
  A a;
  a.i = source();
  A a2 = a;
  sink(a2.i);
}

But we catch it if A is replaced with a struct with more than 1 field like:

struct A {
  int i, j;
};

One problem with adding a PostUpdateNode to the assignment a.i = source() is that the dataflow library requires a predicate getPreUpdateNode that returns the node before the update to its state. For field updates generating a chi instruction we can pick the total operand of the chi (as this represents the memory allocated to the struct), but without a chi instruction it's definately to define a meaningful result of this predicate. So this PR simply implements this predicate as none().

I don't know if this breaks something in the dataflow library, but it does catch the above flow even when A contains a single field.

This PR also adds an additional rule to simpleInstructionLocalFlowStep in order to propagate dataflow to loads with inexact memory operands that arise from stores to single field structs. This rule could be avoided if the overlap was considered MustExactlyOverlap.

[MathiasVP]

Here's the relevant CPP-differences: https://jenkins.internal.semmle.com/job/Changes/job/CPP-Differences/1051/

I'll investigate what's up with boost, drogon and json. There are no result changes, which I guess is unsurprising (I wouldn't expect many real-world cases of structs with a single field).

[MathiasVP]

CPP-differences looks good now: https://jenkins.internal.semmle.com/job/Changes/job/CPP-Differences/1056/

[jbj]

This seems like an improvement overall, so I'll merge.