Prevent shortened label collisions in generated links

[sylvesterkaczmarek]

Fixes #31.

Distinct labels can collapse to the same interface/network name after label shortening (high and hi both become hi). In csv_to_compose.py that reuses the same MultiDiGraph edge key and silently drops one link.

Detect shortened-label collisions per node pair and fall back to the original label only for those collisions. Non-colliding generated names keep the existing shortening behavior.

Validation:

• reproduced the issue on main: two a -> b links labelled high and hi generated one edge
• after the fix the same input generates two distinct edges: a_b_high and a_b_hi
• pytest tests/test_csv_label_collisions.py -q — 2 passed
• Python compile checks pass
• git diff --check passes

[axodentally]

Thanks for the contribution!
Though two links being different, but having the suffix 'high' and 'hi', with all else being equal, is not really a realistic scenario. The conversion scripts are mainly to convert files in the CCSDS scenario reference format, which for the foreseeable future will only be the ones in the upstream repo anyway: https://github.com/esa/ccsds-dtn-reference-scenarios
And those don't have that.

On the other hand it also does not really hurt to include the change since it has no negative consequences. @gh0st42 needs to decide on that.

[sylvesterkaczmarek]

Thanks @axodentally. I agree this is a defensive edge case rather than something present in the current reference scenarios. I have kept the change narrowly scoped and rebased it onto current main; it is now 0 behind and mergeable. Happy to leave the merge decision to @gh0st42.