fix: classify LIMIT sets from name prefixes (release 0.0.3)

[doriannez]

Fixes the set classification observed when importing an Aster-converted
model into LIMIT, compared against the reference Abaqus model of the same
structure.

Two distinct root causes:

1. clean_name() stripped every underscore, so the prefixes LIMIT matches
   literally (PROF_, SW_) were destroyed and their sets landed in
   Other Nsets / Other Elsets instead of Profile Sets and Solid Weld
   Generation Elsets.
2. _write_sections() wrote a *Section for every active elset, and an
   elset with a section becomes a Property Set on import. The industrial
   model came in with 35 Property Sets against 11 on the Abaqus side.
   Only solset* / surfset* groups actually carry a material/thickness.

Both fixes fall back to the previous behaviour when the convention is
absent, so models that do not follow it are unaffected.

Also adds --limit-prefixes / --property-prefixes to override both
conventions, and a warning listing the elsets left without a *Section
(otherwise they reach LIMIT with no material or thickness, unnoticed).