Adding methods to existing functions in test setups is not isolated to the users of the setup

[Drvi]

This led to at least one confusing situation where removing a test setup from a test item didn't cause a failure because the required methods leaked the test setup and were available to the test item regardless, MRE:

@testsetup module B
    import ReTestItems
    ReTestItems.runtests(::Int) = 1
end

@testitem "t1" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t2" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t3" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t4" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t5" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t50" setup=[B] begin @test B.ReTestItems.runtests(1) == 1 end
@testitem "t96" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t97" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t98" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t99" begin @test ReTestItems.runtests(1) == 1 end
@testitem "t100" begin @test ReTestItems.runtests(1) == 1 end

Test Summary:    | Pass  Error  Total  Time
ReTestItems      |    6      5     11  3.4s
  .              |    6      5     11      
    the-tests.jl |    6      5     11      
      t1         |           1      1  0.9s
      t2         |           1      1  0.0s
      t3         |           1      1  0.0s
      t4         |           1      1  0.0s
      t5         |           1      1  0.0s
      t50        |    1             1  0.1s
      t96        |    1             1  0.0s
      t97        |    1             1  0.0s
      t98        |    1             1  0.0s
      t99        |    1             1  0.0s
      t100       |    1             1  0.0s

There isn't probably much to do about this, but maybe we should investigate whether Base.Experimental.@MethodTable could be used to make test setups more isolated.