Make DATAPATH configurable and fix path joins (fixes #3)

[mkzung]

Problem

preprocess_glass.py and preprocess_sub2vec.py hardcode DATAPATH = "./dataset/" and the README tells users to edit it by hand (issue #3). Paths are built by string concatenation with an inconsistent leading slash (DATAPATH+"/background_nodes.csv" vs DATAPATH+"connected_components.csv"), so both groups only work because the default value ends in /. If a user sets the natural DATAPATH="./dataset" (no trailing slash), the second group produces ./datasetconnected_components.csv and the scripts crash with FileNotFoundError.

Fix

• Add a --datapath argument (falls back to $ELLIPTIC2_DATAPATH, default ./dataset) so the folder no longer needs hand-editing (closes Ease of use #3).
• Build every path with os.path.join, removing the trailing-slash dependency and the ./dataset//... double slash.
• Fix a copy-paste bug in the out-of-range warning (n2id[c1] -> n2id[c2] in the c2 branch).
• Correct the README: the split is fixed by the deterministic per-subgraph assignment, not a user-editable "split size" (the unused train/val variables are removed).

Verified without downloading the full dataset: both ./dataset and ./dataset/ now resolve correctly, and both scripts compile.