Update documentation, add tutorial

[muellren]

The current markdown documentation does cover all new features.

• The DeviceArray constructor function
• Namespaces and the CU root object
• Object Shredding

There is usually more than one way to do something in grCUDA. The documentation should be revised to emphasis one canonical way as a form a best practices. Code examples that use features that are planned for deprecation, e.g., array allocation expressions, should be removed.

Add a tutorial-like document that explains

1. how to call a GPU-accelerated host-function in a shared library.
2. how to call a kernel that is defined in an existing cubin or PTX file.
3. how to build a kernel at runtime from a source code string.
4. how to use grCUDA from the various GraalVM languages.