fix(plugins): handle unsigned integer columns in pandas/polars converters

[belowzeroff]

Problem

Unsigned integer columns (uint8/uint16/uint32/uint64) were absent from the pandas and polars dtype maps, so they fell through to Symbol and crashed the converter on otherwise-valid input:

>>> from_pandas(pd.DataFrame({"a": np.array([1, 2, 3], dtype="uint64")}))
TypeError: bad argument type for built-in operation

The same happened for polars UInt8/UInt16/UInt32/UInt64. The pyarrow plugin already handles unsigned columns (#18); pandas and polars were still broken.

Fix

Map unsigned columns to a signed type wide enough to hold every value without wrapping, mirroring the pyarrow plugin:

source dtype	ray type
uint8	U8
uint16	I32
uint32	I64
uint64	I64

uint64 has no lossless target, so values >= 2**63 overflow — identical to pyarrow's behaviour. The numpy "u" dtype kind is also added to the pandas kind fallback.

Tests

Added test_from_pandas_unsigned_ints and test_from_polars_unsigned_ints, each covering all four widths with values above each signed type's midpoint (e.g. uint16 65535, uint32 4294967295) to prove they no longer wrap negative.

pytest tests/plugins/test_pandas.py tests/plugins/test_polars.py  ->  36 passed

🤖 Generated with Claude Code