Added four import functions to pycleora/io_utils.py:

1. from_pandas(df, source_col, target_col, weight_col=None, columns=...)
   - Creates SparseMatrix from a pandas DataFrame
   - weight_col filters out edges with NaN or zero weights
   - NaN filtering uses pd.isna() for pandas nullable dtype safety
   - try/except import pattern for pandas as optional dependency

2. from_scipy_sparse(matrix, entity_ids=None, columns=...)
   - Creates SparseMatrix from scipy sparse adjacency matrix
   - Uses integer string IDs when entity_ids is None
   - Deduplicates symmetric edges via min/max key pattern
   - try/except import pattern for scipy as optional dependency

3. from_edge_list(edges, columns=...)
   - Creates SparseMatrix from list of (source, target) or (source, target, weight) tuples
   - Weight values accepted but not used in SparseMatrix construction
     (actual weighting handled by embed_weighted per architecture)

4. from_numpy(adjacency_matrix, entity_ids=None, columns=...)
   - Creates SparseMatrix from dense numpy adjacency matrix
   - Handles asymmetric matrices by checking both (i,j) and (j,i)
   - Uses integer string IDs when entity_ids is None

All functions use SparseMatrix.from_iterator() internally, follow existing
patterns (from_networkx), include proper error messages and input validation,
and accept standard columns, hyperedge_trim_n, num_workers parameters.

Replit-Task-Id: ceadb42f-60cc-403a-aaa8-2cc81ff94db6