[Core] Implement and align np typing functions

[Nucs]

Overview

Implement NumPy's type introspection and promotion functions (np.iinfo, np.finfo, np.can_cast, etc.) and consolidate NumSharp's scattered type information into a centralized, NumPy-aligned system.

Problem

NumSharp's type information is scattered across 5+ files with duplicates and inconsistencies:

File	What it has
NPTypeCode.cs	GetDefaultValue(), SizeOf(), AsType(), IsUnsigned(), IsSigned(), GetAccumulatingType()
NumberInfo.cs	MinValue(), MaxValue()
TypeRules.cs	GetTypeSize(), GetClrType(), IsUnsigned(), IsSigned(), IsFloatingPoint(), GetAccumulatingType()
InfoOf<T>	Size, Zero, MaxValue, MinValue (generic cached)
ReductionTypeExtensions	GetOneValue(), GetMinValue(), GetMaxValue() (uses infinity for floats)

Duplicates found:

• SizeOf() vs GetTypeSize() vs InfoOf<T>.Size (3 implementations)
• AsType() vs GetClrType() (2 implementations)
• IsUnsigned(), IsSigned() (2 implementations each)
• GetAccumulatingType() (2 implementations)
• MinValue()/MaxValue() vs GetMinValue()/GetMaxValue() (different semantics!)

NumPy has clean, unified APIs (np.iinfo, np.finfo) that NumSharp lacks entirely.

Proposal

Phase 1: Implement np.iinfo and np.finfo

• Create np.iinfo(int_type) class returning:

  • bits - number of bits (8, 16, 32, 64)
  • min - smallest value
  • max - largest value
  • dtype - the NPTypeCode
  • kind - 'i' (signed) or 'u' (unsigned)

• Create np.finfo(float_type) class returning:

  • bits - number of bits
  • min, max - value range
  • eps - machine epsilon
  • epsneg - negative epsilon
  • tiny / smallest_normal - smallest positive normal
  • smallest_subnormal - smallest positive subnormal
  • precision - decimal digits
  • resolution - 10^-precision
  • maxexp, minexp - exponent limits
  • dtype - the NPTypeCode

Phase 2: Implement Type Checking Functions

• np.issubdtype(arg1, arg2) - check dtype inheritance
• np.isdtype(dtype, kind) - check dtype against kind string
• np.issctype(rep) - check if scalar type
• np.issubsctype(arg1, arg2) - check scalar type inheritance
• np.sctype2char(sctype) - scalar type to character code
• np.maximum_sctype(t) - highest precision of same kind

Phase 3: Implement Type Promotion Functions

• np.can_cast(from, to, casting="safe") - check if cast is valid
• np.result_type(*arrays_and_dtypes) - common result type (make public)
• np.promote_types(type1, type2) - smallest type to hold both
• np.min_scalar_type(a) - minimum type to hold scalar
• np.common_type(*arrays) - common type for arrays

Phase 4: Consolidate Internal Type Info

• Create centralized NPTypeInfo static class with all type metadata
• Remove duplicate GetTypeSize() from TypeRules.cs (use NPTypeCode.SizeOf())
• Remove duplicate GetClrType() from TypeRules.cs (use NPTypeCode.AsType())
• Remove duplicate IsUnsigned(), IsSigned() from TypeRules.cs
• Remove duplicate GetAccumulatingType() from TypeRules.cs
• Move ReductionTypeExtensions.GetOneValue() to NPTypeCode extensions
• Keep ReductionTypeExtensions.GetMinValue()/GetMaxValue() separate (uses infinity - different semantics)
• Update all callers to use consolidated APIs

Phase 5: Additional Type Functions

• np.isreal(x) - check if array has no imaginary part
• np.iscomplex(x) - check if array has imaginary part
• np.iscomplexobj(x) - check if object is complex type
• np.isrealobj(x) - check if object is real type

NumPy vs NumSharp Current State

NumPy Function	NumSharp Status
np.iinfo(dtype)	❌ Missing
np.finfo(dtype)	❌ Missing
np.isscalar(x)	✅ Implemented
np.issctype(rep)	❌ Missing
np.issubdtype(arg1, arg2)	❌ Missing
np.isdtype(dtype, kind)	❌ Missing
np.result_type(*arrays)	⚠️ Internal only (_FindCommonType)
np.find_common_type(array_types, scalar_types)	✅ Implemented
np.promote_types(t1, t2)	❌ Missing (internal only)
np.can_cast(from, to)	❌ Missing
np.min_scalar_type(a)	❌ Missing
np.common_type(*arrays)	❌ Missing
np.isnan(x)	✅ Implemented
np.isinf(x)	✅ Implemented
np.isfinite(x)	✅ Implemented
np.isclose(a, b)	✅ Implemented
np.isreal(x)	❌ Missing
np.iscomplex(x)	❌ Missing
np.sctype2char(sctype)	❌ Missing
np.maximum_sctype(t)	❌ Missing

Evidence

NumPy source reference: src/numpy/numpy/_core/getlimits.py (iinfo, finfo)
NumPy source reference: src/numpy/numpy/_core/numerictypes.py (type functions)
NumPy source reference: src/numpy/numpy/_core/multiarray.py (can_cast, result_type, min_scalar_type)

Scope / Non-goals

In scope:

• Type introspection classes (iinfo, finfo)
• Type checking functions (issubdtype, isdtype, etc.)
• Type promotion functions (can_cast, promote_types, etc.)
• Consolidating scattered internal type info
• Removing duplicates

Out of scope:

• Complex number support (separate issue)
• String dtype support (see [Core] NumPy String support in NumSharp #592)
• datetime64/timedelta64 support