fix: return Result from draw() and save_gexf() (closes #33)

[Dani-giron]

Summary

draw() exports the HNSW graph to GEXF files and had two silent failure modes. gexf.to_string() can fail and was called through .unwrap(), which panics the whole process on a serialization error instead of surfacing it. Separately, draw() called save_gexf() through let _ = ..., discarding its std::io::Result and swallowing I/O errors from fs::write (disk full, no permission, invalid path) while still returning as if the export had succeeded.

Changes

• src/controllers/apotheosis.rs: both draw() and the private save_gexf() now return Result<(), Box<dyn std::error::Error>>. The to_string() calls use ? instead of .unwrap(), and the save_gexf() call in draw() uses ? instead of let _ =. GexfError from the gexf crate already implements std::error::Error via thiserror, so it converts into Box<dyn std::error::Error> through ? with no extra work.
• This is a public signature change. No caller in src/ or src/bin/ calls draw(), so nothing there needed updating.
• tests/api_contract.rs: draw_produces_one_gexf_file_per_layer, the one test calling draw(), now calls .expect() on the result.

Test plan

Verified locally with the same commands CI runs: cargo clippy --all-targets --all-features -- -D warnings clean, cargo fmt --check clean, and all 44 tests passing.

Also reproduced the silent I/O failure live before fixing it: calling draw() with a path whose parent directory does not exist returned Ok with no panic and no file written, confirming the bug was real and not just a static-analysis finding.

Closes #33.

[Dani-giron]

Heads up: this PR will conflict with #25 for the same reason, see the comment there. Whichever merges second needs to either reapply this fix at the new location (src/export/gexf.rs) or rebase around #25's move, depending on order.