`__hash__`, `__eq__`, ... should not require a CUDA context

[leofang]

Example:

cuda-python/cuda_core/cuda/core/_stream.pyx

Lines 183 to 186 in 39e3293

	def __hash__(self) -> int:
	# Ensure context is initialized for hash consistency
	Stream_ensure_ctx(self)
	return hash((as_intptr(self._h_context), as_intptr(self._h_stream)))

@stiepan reported offline that this is problematic because at least for streams we do not always know the CUDA context associated with them (we could be wrapping a foreign stream), and triggering a lookup is expensive. Some of the objects are even independent of CUDA contexts.

We decided to follow cccl-rt and do a simple pointer comparison for __eq__. We should stick to this principle for __hash__ as well.