Extends gt, ge, lt, le, and multiple_of Meta constraints to work with
decimal.Decimal types, enabling exact arithmetic comparisons and
sub-integer precision constraints without floating point issues.

Constraint bounds passed as int or float are coerced to Decimal via
their string representation to preserve precision. multiple_of uses
Python's % operator for exact arithmetic.

Closes #683

Coercing a Decimal bound to int or float silently drops precision, which
defeats the point of specifying the bound as a Decimal in the first
place. Only allow Decimal bounds when the annotated type is also
Decimal, and raise a TypeError otherwise at TypeNode construction time.

# Conflicts:
#	src/msgspec/__init__.pyi
#	tests/unit/test_constraints.py

Frees up 4 bits in the types bitmask (main now uses bits 38-39 for
frozendict, leaving too few free bits for per-constraint decimal bits).
Also compare the multiple_of remainder against zero explicitly instead
of relying on __bool__.

MS_TYPE_FROZENDICT took bits 38-39, MS_CONSTR_DECIMAL moved to bit 40.

# Conflicts:
#	src/msgspec/__init__.pyi

…string

The float branch already rejected inf and nan bounds; the Decimal branch
converted to double for the positivity check only and let Decimal('NaN')
and Decimal('Infinity') through. Also check PyObject_IsInstance for error.