C++: Repair MustFlow library for use-use flow

[MathiasVP]

The MustFlow library was previously defined in terms of the IR-based dataflow nodes, and implicitly relied on the assumption that all relevant instructions had dataflow nodes.

On the feature branch, this is no longer the case. Thus, we cannot use the dataflow nodes, and we have to use the IR directly.

Sadly, this is a breaking change. I think this is okay, however, because:

1. I haven't seen any uses of it outside the two queries we have in Code Scanning that use it.
2. It's really easy to transition from the dataflow nodes to the instructions (since the library was already using the IR-based dataflow library).

[jketema]

I wonder if this PR should just target main instead of the feature branch, as it's pretty independent of the use-use dataflow changes.

[MathiasVP]

I wonder if this PR should just target main instead of the feature branch, as it's pretty independent of the use-use dataflow changes.

Indeed, I'd also be okay with this.

[MathiasVP]

I've retargeted the PR to target main now. I managed to do so without pinging all teams 🎉, but I didn't manage to stop the bot from adding a bunch of unnecessary labels 😭. Oh well!

[jketema]

One typo and a number of related questions. The latter boil down to: why can we get away with just Instructions as sources and Operands as sinks?

[jketema]

We should probably also do a DCA experiment once my questions have been resolved.

[MathiasVP]

We should probably also do a DCA experiment once my questions have been resolved.

Good point. I've started a run now.

[jketema]

I'm happy if DCA is happy.

[MathiasVP]

DCA looks fine 🎉