Fix dispatch for rdiv! with LU

[dkarrasch]

This method (added in #31853) was more specific than the one on line 493, which was added in #31285 at about the same time.

Fixes #55422.

[dkarrasch]

This will need to be backported manually...

[dkarrasch]

For the reproducer in the original issue, I get

julia> A = rand(1000,1000); @btime inv($A); @btime inv2($A); @btime inv3($A);
  25.634 ms (10 allocations: 8.13 MiB)
  38.917 ms (9 allocations: 15.27 MiB)
  37.542 ms (9 allocations: 15.27 MiB)

julia> inv(A) ≈ inv2(A) ≈ inv3(A)
true

[dkarrasch]

@jishnub The failures are unrelated, but still concerning. Do you remember if those NaN-ignoring things were known and maybe fixed? Since these tests are random, they may become invisible in another CI run, but we need to handle them, right?

[jishnub]

I think these were fixed in #55737 and #55777.

The best way to check would be to rebase and rerun the tests locally by specifying the same random seed that had caused the failures, as suggested in #55727 (comment)

[dkarrasch]

How do I extract that seed from "my" previous run? Is it printed somewhere in the test logs? Oh, or is it the "global RNG seed" printed after the big table?

[jishnub]

Yes, that's the seed

[dkarrasch]

Okay, unfortunately it failed only on Mac, and not on linux, so I can't test it locally. Anyway, let's merge this when tests pass, and assume this was nothing new, but the fixed issue.

[jishnub]

But the tests on Linux and Mac must have used different seeds, no? Tests for a specific seed would fall on all platforms if it's due to the issue above. In any case, I think it's fixed now.