Backports for 1.6, take 2.

[KristofferC]

No description provided.

[IanButterworth]

I tested this locally against JuliaLang/julia#40702 and got the same error

undo redo functionality: Test Failed at /Users/ian/Documents/GitHub/Pkg.jl/test/pkg.jl:700
  Expression: !(haskey(Pkg.dependencies(), unicode_uuid))

No clue..

[IanButterworth]

@KristofferC I've looked at this a few times and still don't have any idea what broke the undo tests

[IanButterworth]

2018f15 is the first bad commit for the undo tests, testing with JuliaLang/julia#40702

Also, 27ab070 seems like a bad backport as Pkg.upgrade_manifest() wasn't backported to 1.6 but that backport add a test for it, so that'll fail.
I don't think what was in that original PR is needed at all as there are later commits that fix the other changes it made. So I'd just drop it.

[IanButterworth]

I think the problem is that the manifest equality is failing here https://github.com/JuliaLang/Pkg.jl/blob/kc/1.6/src/API.jl#L1539

I guess during the undo tests the original_manifest that's deepcopied during the intial Context() is missing the julia_version which is added in the initial resolve during the first Pkg.add, so every comparison afterwards back to that will result in ctx.manifest == ctx.original_manifest being false, which means the undo snapshot will always be added.

Perhaps == for manifests should just check the deps?
PR: #2649

Though, I don't yet understand why the undo tests aren't failing on master..

Edit: On second thought.. I'm really not certain I've got a handle on this properly. When original_manifest is updated throughout the EnvCaching is confusing me

Edit edit: Seems like my suggestion passes #2650 but still not clear why it didn't fail on master

[IanButterworth]

The fix agreed with @KristofferC was to change the undo functionality to just consider equality wrt. the manifest deps, rather than change the equality method for the manifest type to do that

[DilumAluthge]

@KristofferC @fredrikekre Is this good to go?