Adjust margin of ARIMA flaky test

[dantegd]

No description provided.

[vyasr]

I'm confused why there are so many commits in this PR, but the diff is one line and all the intermediate commits will get squashed out so this seems OK to me.

[dantegd]

I've used the same branch for all the fixes for Scipy 1.15, since we have squash commits then the PRs show all the commits I've done :)

[betatim]

There is also a TSNE test that just fails (assert 0.84 < 0.85). Might resolve itself with rerunning.

Fine with increasing the tolerance for arima if there is nothing else we can do