Profile and optimize slow GraphML export in `tsumugi build-graphml`

[akikuno]

📋 Description

The GraphML export in TSUMUGI is slow.

tsumugi build-graphml takes too long to finish, and the bottleneck may be in the GraphML writing step.

Problem

• The execution time of tsumugi build-graphml is too long.
• graphml_builder.py should be profiled to identify the bottleneck.
• networkx.write_graphml may be the main cause of the slowdown.

Proposed actions

1. Profile the execution speed of graphml_builder.py

   • Measure total runtime
   • Identify the slowest functions/steps
   • Confirm whether GraphML serialization/writing is the bottleneck

2. Evaluate networkx.write_graphml

   • Verify how much time is spent in networkx.write_graphml
   • Check whether performance degrades with larger graphs

3. Consider replacing the backend if needed

   • If networkx.write_graphml is confirmed to be too slow, consider replacing NetworkX-based GraphML export with a faster implementation (e.g., rustworkx)

Expected outcome

• Clear identification of the performance bottleneck in GraphML export
• A concrete optimization plan (code-level improvements and/or backend replacement)
• Improved runtime for tsumugi build-graphml

🔖 TSUMUGI Version

No response

📎 Anything else?

Thanks @kinari-labwork for reporting the issue!