Avoid comparing window range canonicalized plans on Spark 3.0.x

[jlowe]

Fixes #2964

The logical optimizer in Spark 3.0.x is not deterministic when it comes to the processing order of windows that have the same ranges. Therefore on Spark 3.0.x it doesn't make sense to compare canonicalized physical plans for these cases since the logical plans from which the physical plans were derived are not guaranteed to be the same.

[jlowe]

@mythrocks I noticed there's a ton of commented-out tests WindowFunctionSuite stating they are waiting on #1039 which is now fixed. However it also mentions that these are mostly redundant with the Python integration tests.

I went ahead and updated the commented code in this PR, but I'm wondering what we should do with these disabled tests going forward (potentially in a separate PR). Should these be enabled, or are they providing no value beyond what the integration tests are already covering?

[jlowe]

build

[mythrocks]

@mythrocks I noticed there's a ton of commented-out tests WindowFunctionSuite stating they are waiting on #1039 which is now fixed. However it also mentions that these are mostly redundant with the Python integration tests.

Apologies for the delayed response. We can deal with those tests in a follow-up issue. My view is that those can be removed, since we're already testing the null-ORDER BY column case in window_function_test.py.