Conform floor division and modulo to Python semantics

[Viriathus1]

This PR correctly implements floor division (//) and modulo (%) operators, fully matching Python’s semantics for both integers and floats.

Changes include:

• Signed Integer floor, eg:

int32_t q = x / y;
int32_t r = x % y;

if ((r != 0) && ((x ^ y) < 0)) {
    q -= 1;
}

return q;

Now it uses pure integer arithmetic which matches CPython's internal implementation and ensures negative values round correctly toward negative infinity.

• Signed Integer modulo, eg:

int32_t r = x % y;

if ((r != 0) && ((x ^ y) < 0)) {
    r += y;
}

return r;

Guarantees that the result of x % y always has the same sign as y consistent with Python.

• Added float modulo operator.
• Added new unit tests to cover all combinations of positive and negative dividends and divisors.

This closes #238

[antocuni]

do you have any specific reason for killing this?
I personally find having Annotated[W_IntLike, w_T] everywhere hard to read, and this patterns is used a bit everywhere, e.g. here:

spy/spy/vm/member.py

Lines 61 to 78 in ead6d95

	@builtin_method('__get__', color='blue', kind='metafunc')
	@staticmethod
	def w_GET(
	vm: 'SPyVM', wam_self: 'W_MetaArg', wam_obj: 'W_MetaArg'
	) -> 'W_OpSpec':
	from spy.vm.opspec import W_OpSpec
	w_self = wam_self.w_blueval
	assert isinstance(w_self, W_Member)
	w_T = wam_obj.w_static_T
	field = w_self.field # the interp-level name of the attr (e.g, 'w_x')
	T = Annotated[W_Object, w_T] # type of the object
	V = Annotated[W_Object, w_self.w_type] # type of the attribute
	@vm.register_builtin_func(w_T.fqn, f"__get_{w_self.name}__")
	def w_get(vm: 'SPyVM', w_obj: T) -> V:
	return getattr(w_obj, field)
	return W_OpSpec(w_get, [wam_obj])

[Viriathus1]

@antocuni I was running up against Pylance Type error Variable not allowed in type expression which made it hard to work with. I've just changed the language server to Jedi and it works ok with that pattern so I'm happy to revert it.

It could be good to have the recommended static type checker (mypy) and language server on the readme for local development because they're all slightly different and the default language server on vscode is Pylance.

[antocuni]

I recommend mypy: this is what the CI tests and it's the only one "officially" supported.
I'm not opposed to switch to another type checker, but to do so we would need to understand pros&cons and to test it in CI.

Ironically, the fact that in Python "type checking" is not a well defined concept and that every typechecker implements slightly different rules is exactly one of the things that SPy aims to solve :).
But for that, we need to wait until SPy is powerful enough to compile itself :)

[antocuni]

LGTM, apart the asserts