chore: some Jarl fixes

[maelle]

Fix #2867

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if eb00542 is merged into main:

• ✔️as_adjacency_matrix: 744ms -> 742ms [-1.07%, +0.67%]
• ✔️as_biadjacency_matrix: 744ms -> 741ms [-1.01%, +0.42%]
• ✔️as_data_frame_both: 1.61ms -> 1.61ms [-1.17%, +1.15%]
• ✔️as_long_data_frame: 4.02ms -> 4.03ms [-1.76%, +2.07%]
• ✔️es_attr_filter: 2.81ms -> 2.83ms [-2.89%, +4.74%]
• 🚀graph_from_adjacency_matrix: 117ms -> 116ms [-2.16%, -0.26%]
• ✔️graph_from_data_frame: 3.76ms -> 3.72ms [-2.3%, +0.15%]
• ✔️vs_attr_filter: 1.61ms -> 1.61ms [-4.03%, +3.96%]
• ✔️vs_by_name: 1.01ms -> 1ms [-4.08%, +2.69%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[schochastics]

LGTM