What does this implement/fix? Explain your changes.

Passing a non-integral numpy float where a .NET integer parameter is expected silently truncates instead of raising: SimpleMovingAverage(np.float64(20.5)) builds a period-20 indicator, while plain float(20.5) is correctly rejected with a TypeError.

fb5cce6 (#128) added the non-integral rejection for Python floats, but the guard in Converter.ToPrimitive uses Runtime.PyFloat_Check, an exact type-pointer check. Two kinds of values bypass it and fall through to PyNumber_Long/__int__, which truncates:

• float subclasses — numpy.float64 is a subclass of Python float
• numbers that only define __float__ — numpy.float32/float16 are not float subclasses

This PR extends the guard to any float-like value:

• a Python float including subclasses (PyObject_TypeCheck against PyFloatType), or
• a number defining __float__ but not __index__ (Python's marker for losslessly-int-convertible types)

Integral-valued floats (np.float64(20.0)) keep converting, numpy integer scalars (np.int64, np.int32, which advertise __index__) are unaffected, and plain ints take a fast early exit. If a probed __float__ call fails, the pending Python error is cleared before rejecting, so no stale error leaks.

Behavior note: non-numpy numeric types with __float__/__int__ but no __index__ (e.g. decimal.Decimal, fractions.Fraction) previously also truncated silently into integer parameters (Decimal("20.5") → 20); with this change their non-integral values are rejected the same way, while integral values keep converting exactly as before.

Does this close any currently open issues?

No open issue; found while reproducing the fleet error-surface study (Agents#305, improvement 1).

Any other comments?

Tests:

• TestFloatToIntConversion embed tests: new float-subclass, __float__-only and __index__ fixture cases over single and overloaded targets (22 pass).
• tests/test_conversion.py::test_numpy_float_to_int_conversion: numpy-backed coverage (np.float64/np.float32 integral and non-integral, np.int32/np.int64, plain-float regression, method binding), skipped automatically when numpy is absent.
• Full embed suite: 982 passed, 0 failed. Full python suite: 457 passed; the single failure (test_module.py::test_explicit_assembly_load) is a local-environment artifact and fails identically on clean master.

Checklist

Check all those that are applicable and complete.

• Make sure to include one or more tests for your change
• If an enhancement PR, please create docs and at best an example
• Ensure you have signed the .NET Foundation CLA
• Add yourself to AUTHORS
• Updated the CHANGELOG