PyPy raise on known errors

[milesgranger]

Part of #142.

Raise on errors known to fail with PyPy.

• No bytes or memoryview for target output (only relavent for de/compress_into methods)
• Will sometimes fail with multidimensional numpy arrays depending on dtype, notably datetime dtypes and a few others.

Additionally this:

• Updates CI to run the tests on PyPy environments
• Update PythonBuffer::as_slice_mut to return a Result so we can raise if we're on PyPy in the context mentioned above for raising errors.
• Better checks from C api for checks comparing to 1 for true.
• Drops PyPy 3.7 and 3.8, adding 3.10 and keeping 3.9.
• Updates Cargo.lock
• Parametrized pypy builds for macos and linux, Windows fails but that's a bigger issue