TD3: Eliminate Some Recursive Toxic Trash (namely the directly recursive one)

[michael-schwarz]

@jerhard and I realized that implementing the removal of directly recursive trash is quite easy. This PR adds this and a test where it makes a difference.

This is against a new branch on the same state as #1636, so the PR appears in this repo and not in the fork. Once #1636 is merged (hopefully soon), we can change the base for this.

[michael-schwarz]

I made a fork also of the experiments repo: https://github.com/tum-cit-pl/precision-recovery-mixed-flowsens-benchmarks/tree/cyclic

For now, there is no additional precision gain for large programs: plot4.pdf

[michael-schwarz]

There is one program where we discover one additional call in a context to be dead:

Warning: siege.c has different dead counts for ours-bot-rec and ours-bot
Dead with ours-bot-rec: (756, 756, 808, 3198)
Dead with ours-bot: (755, 755, 808, 3198)

where the format is stillbot,botified,dead,total.

The experiments are run on https://github.com/goblint/analyzer/tree/pldi25_eval_stats_rectrash.

[sim642]

What does "directly recursive" mean? A function in a context depending on the same function in the same context, and only that?

[michael-schwarz]

A function in a context depending on the same function in the same context, and only that?

Yes!

[michael-schwarz]

Even after the fixes by @jerhard and me, there is only one program in the large subset where this makes a difference.

[michael-schwarz]

It does not seem like it would make sense to pursue this for now.