[BUG] hash join gather maps limited by size_type

[revans2]

Describe the bug
Hash join gather maps were implemented to get around issues when a join explodes. #7454

But there is still a limit in place on the maximum result size that can be returned by a join.

cudf/cpp/src/join/hash_join.cuh

Lines 152 to 154 in cea6c20

	CUDF_EXPECTS(h_size_estimate <
	static_cast<estimate_size_type>(std::numeric_limits<cudf::size_type>::max()),
	"Maximum join output size exceeded");

Steps/Code to reproduce bug
I was trying to probe the limits of what the new Spark out of core join implementation could do, and I hit this when trying to join a lot by a lot. 10 billion rows on the left hand size and 10 million on the right hand side. But because of how the joins work in spark it was a shuffled hash join so there were 4 separate joins happening. The keys were evenly distributed and there were 100,000 of them, so you could probably divide all the numbers by 4 to get the error, but it should not be hard to do.

Expected behavior
We can produce a gather map that is larger than can fir in a column.

[shwina]

Good catch. Looks like we need to use size_t here instead of size_type. Fix coming up.