Allow more control over argument conversion

We currently extract all arguments using the FromPyObject trait and often run into conflicts whether this should be strict, i.e. basically only downcasting the Python objects to the Python equivalents of the target Rust types, or loose, i.e. using the most general available Python API to produce an instance the target Rust type even if these are conceptually unrelated. Concrete example as are PyBool/bool and whether this should use only downcasting or the full truth value testing procedure or PySequence/Vec which might check whether the Python type is actually or sequence or take any iterable and collect it into a Rust Vec.

One way to alleviate this issue is to allow the user to choose which strategy to pursue, for example by adding a #[pyo3(noconvert)] argument attribute to explicitly request a strict conversion.

Start by tracing how the FromPyObject trait handles function arguments and how argument attributes are parsed. Compare strict downcasting with the broader conversion behavior described for PyBool/bool and PySequence/Vec. Done means users can explicitly request strict conversion, such as with a #[pyo3(noconvert)] argument attribute, without removing existing conversion behavior.