Provide better diagnostics for the debug mode

[steve-s]

At this point, if HPy detects an issue in the debug mode, it prints a generic message and aborts the process(*).

It should provide more context to help with debugging of the issue, i.e., "diagnostics output":

• info about the underlying Python object to help to identify it, like inspect: type, length for lists, tuples, etc., or maybe just call __repr__?
• stack trace of the location where the handle was allocated
• stack trace of the location where the issue was detected
• anything else?

Implementation:

• retaining the stack trace of the allocation can be expensive, so I suggest it can be configured, it will be off by default, but the diagnostics output will give instructions how to turn it on
• we need a way to get a stack trace
  • roll out our own (simple) implementation, on MacOS and Linux based on backtrace, there seems to be some API for this on Windows too
  • use 3rd party library:
  • unwind
  • support in boost

(*) In case of the mmap based raw data protection mechanism (#259), it just segfaults, which we should improve by registering a handler and then we can provide the same diagnostics there too.

[steve-s]

We can start by defining some simple internal abstraction: report_invalid_handle(handle), and some basic implementation that would use backtrace where available and print some generic message otherwise. This should be fairly simple and would already provide some nice improvement.