[BUG] Inconsistent Java points-to graph and grammar conventions across 21 datasets

[homka122]

Describe the bug

The 21 Java datasets contain 42 java_points_to*.cnf files, but fall into two groups:

Datasets	Grammar templates	Downloaded graphs
14: avrora, batik, eclipse, fop, h2, jython, luindex, lusearch, pmd, sunflow, tomcat, tradebeans, tradesoap, xalan	Use load/store and unindexed intermediate nonterminals	Indexed forward edges; no reverse edges
7: commons_io, commons_lang3, gson, guava, jackson, junit5, mockito	Use _i for indexed terminals and intermediate nonterminals	Indexed forward edges and already-added reverse edges

The documentation assigns reverse-edge augmentation to the user. The absence of reverse edges in the first group is expected. Their presence in the second group means that applying cfpq_data.add_reverse_edges() also reverses edges that are already reverse edges. For example, a graph containing load_2 and load_r_2 gains load_2_r and load_r_2_r.

To Reproduce

1. With CFPQ_Data 5.0.0, download avrora and gson using cfpq_data.download(...).
2. Inspect their graph labels and both supplied java_points_to*.cnf files.
3. Observe that avrora has only forward edges, while gson already has reverse edges such as load_r_2.
4. Apply cfpq_data.add_reverse_edges() to the gson graph and inspect the resulting labels.

Expected behavior

Synchronize the Java graph archives, grammar conventions, and documentation. If users are expected to add reverse edges, downloaded graphs should consistently contain only forward edges, so augmentation is applied exactly once. Document how both grammar-template notations are materialized and which reverse-label naming convention should be used.

Desktop

• OS: Fedora Linux 44 (Sway), x86_64
• CPU: AMD Ryzen 5 5500U
• RAM: 14 GiB
• CFPQ_Data: 5.0.0
• Python in project .venv: 3.13.13

Screenshots