fix: reject negative entries in numpy-array cone dims

[bodono]

Summary

get_cone_arr_dim in scs/scsobject.h handles three input shapes for integer cone fields (q, s, cs, etc.):

• Python list → parse_pos_scs_int (rejects negatives)
• bare scalar → get_pos_int_param → parse_pos_scs_int (rejects negatives)
• numpy array → straight memcpy into scs_int * (no sign check)

The array branch thus let negatives slip past Python-side validation. They'd then surface as a misleading "ScsWork allocation error!" ValueError, with SCS core printing "soc cone dimension error" to stderr before failing workspace init.

Reproduced locally:

>>> scs.SCS(..., {"q": [-1]}, ...)
ValueError: Invalid value for cone field 'q'
>>> scs.SCS(..., {"q": -1}, ...)
ValueError: Invalid value for cone field 'q'
>>> scs.SCS(..., {"q": np.array([-1])}, ...)
soc cone dimension error                          # <-- to stderr, from SCS core
cone validation error
ValueError: ScsWork allocation error!              # <-- wrong & generic

Fix

After the memcpy, walk the copied entries and printErr(key) on any negative — same error shape the list/scalar branches produce.

Tests

Added parametrized tests covering list, scalar, and numpy-array inputs for cone fields q and s, all asserting:

ValueError: Invalid value for cone field '<key>'

Full suite locally: 334 passed, 66 skipped.

Independent of #201 / #202 / #203 — touches only scs/scsobject.h (disjoint from #201) and test/test_scs_coverage.py (disjoint from others).

🤖 Generated with Claude Code