fix: schema.cypher rel and output_dir naming

[adsharma]

What

Two fixes for the vertex/edge Parquet → icebug-disk conversion path.

1. Valid schema.cypher (NODE/REL name clash)

Previously the generated schema used the same table name for both the NODE and REL table:

CREATE NODE TABLE graph500_24(id INT64, PRIMARY KEY(id)) WITH (storage = './tmp', format = 'icebug-disk');
CREATE REL TABLE graph500_24(FROM graph500_24 TO graph500_24) WITH (storage = './tmp', format = 'icebug-disk');

The REL table is now named <graph>_rel, so NODE and REL catalog names never collide:

CREATE REL TABLE graph500_24_rel(FROM graph500_24 TO graph500_24, weight DOUBLE) ...

2. output_db → output_dir

convert_parquet_dir_to_csr had no DuckDB involvement — the output_db path was only a naming convention. It is replaced with output_dir:

• Default output is now <source_dir>-csr (e.g. graph500-24/ → graph500-24-csr/)
• Multi-graph sources write <output_dir>/<graph_name>/
• New --output-dir CLI flag for --source-dir; --output-db remains for the DuckDB/GraphAr paths
• Default storage recorded in schema.cypher derives from the output dir name

Test plan

• uv run pytest: 109 passed
• Manual CLI run on a graph500-24 source dir: output lands in graph500-24-csr/ with a valid schema.cypher