Replace RowDictWithCache with on-demand RowFromId to fix memory exhaustion

[Copilot]

RowDictWithCache eagerly materialised all n^n possible rows at once — for n=8 that's 16,777,216 entries — making it unusable for even moderately large orders.

Changes

• lib/helper.gd

  • Replaced RowDictWithCache with RowFromId(row_id, n): computes a single row on demand using the existing PowersOfNWithCache helper, no pre-allocation.
  • MultiplicationTableDecode: calls RowFromId per row instead of indexing into a precomputed dictionary.
  • MultiplicationTableGetEntry: decodes the column entry directly without constructing any row intermediate, preserving O(1) access.

• tst/test_helper_multiplication_table_encoding.tst: replaced RowDictWithCache tests with RowFromId tests covering row ID 0 (all-ones), max row ID n^n−1 (all-n), and intermediate IDs across multiple values of n.

# Before: builds and caches all n^n rows upfront
row_dict := __SmallAntimagmaHelper.RowDictWithCache(n);  # 16M entries for n=8
return row_dict[row_id + 1];

# After: decodes exactly the one row needed
return __SmallAntimagmaHelper.RowFromId(row_id, n);

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.