Change order of initialization so pinned pool is available for spill framework buffers

[abellina]

While collecting traces for #11888, I realized that the bounce buffers I am using are pageable, which is pretty bad. If I change the order of initialization, the pinned pool can be initialized before the spill framework, so we can then use pinned memory for the d2h and h2d bounced copies.

Renamed init functions slightly, as their names are not matching their function currently.

[abellina]

build

[abellina]

CI failures due to tests initializing too many times. Before my change we had all of this code under a "make sure Rmm is not initialized already" block.

[abellina]

build

[zpuller]

LGTM, was able to confirm we see spill going to pinned mem

[revans2]

I'm not super happy with the TestUtils.withGpuSparkSession needing to be removed. I'm not sure how to tell when reviewing something if it should be there or not. But that said it is minor.

[abellina]

I'm not super happy with the TestUtils.withGpuSparkSession needing to be removed. I'm not sure how to tell when reviewing something if it should be there or not. But that said it is minor.

I don't know how to explain it other than we shouldn't start spark sessions when we don't need to submit spark tasks during a test. I think it's fairly straightforward: if your suite is all about spark tasks submitted via SQL or spark APIs, then yeah by all means set up a session.. but if your tests are really unit tests, don't use the spark session to backwards initialize singletons, just set them up directly.