Avoid collecting TD3 dep when non-incremental

[sim642]

Closes #1566.

I'm wondering if this makes any difference for large programs like rsync.
The biggest hashtables in TD3 (and probably Goblint overall) are rho, infl and dep because they contain most encountered unknowns. So keeping one empty could make quite a difference.

Obviously we should have this anyway, but I'm interested in doing some benchmarking on sv-benchmarks and rsync to find out if this is one of the stupid significant bottlenecks.

TODO

• sv-benchmarks
• rsync

[Copilot]

Pull request overview

Reduces memory overhead in the TD3 solver by avoiding population of the (potentially large) dep dependency table when it isn’t needed for incremental functionality.

Changes:

• Introduces a collect_dep flag to control whether dep is populated.
• Guards dep updates in add_infl and dep resets on RHS re-evaluation behind collect_dep.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sim642]

On sv-benchmarks with level01, 60s and 1GB, it doesn't make a difference in time, but it does lower peak memory usage by ~5%:

(https://goblint.cs.ut.ee/results/322-all-level01-pr-1972-after/table-generator-cmp.table.html#/scatter?toolX=0&columnX=4&toolY=1&columnY=4&regression=Linear&results=All&filter=0(0*status*(status(notIn()),category(notIn(error)))),1(0*status*(status(notIn()),category(notIn(error))))&scaling=Linear&line=1.2)

I still want to try it on rsync to see what happens for significantly larger programs.

[sim642]

This speeds up rsync analysis by ~6% (on top of #1974): goblint/bench@6e4ada7.