Improve the typing of pint wrapped method

[alihamdan]

This PR improves the user facing types of parameters of methods wrapped with pint wrappers. Inside the methods, the types of parameters that accept either a quantity or a unitless value become incorrect but users of these methods now get the correct types. This means we accept a little bit of wrong types in our internal code for the benefit of our users. Until the Python type system have a "Map" type support, this is our best option.

[alihamdan]

Example:

before

@ureg.wraps("V", ("V",), strict=False)
def f(v: float) -> Q_[float]:
    do_something(v)  # OK: v here is float, typed as float.
    ...

# user code
f(230.5)  # OK
f(Q_(20, "kV"))  # WRONG, v accepts float only

after

@ureg.wraps("V", ("V",), strict=False)
def f(v: float | Q_[float]) -> Q_[float]:
    do_something(v)  # WRONG: v here is float, typed as float | Q_[float].
    ...

# user code
f(230.5)  # OK
f(Q_(20, "kV"))  # OK

[alihamdan]

Also numpy 2.0 removes np.float_ and np.complex_ https://numpy.org/devdocs/release/2.0.0-notes.html so I replaced them before they start emitting deprecation warnings