Fate of qr

[martinholters]

From our README:

• Compat.qr takes pivot as a Val instance and keyword argument full ([#22475], [#24279]).

And we have these tests:

Compat.jl/test/runtests.jl

Lines 1676 to 1695 in 2a82221

	let A = [1 2; 1 2; 1 2]
	f = Compat.qr(A, Val(false), full=false)
	@test f == Compat.qr(A, Val(false))
	@test length(f) == 2
	@test size(f[1]) == (3, 2)
	@test f[1] * f[2] ≈ A
	f = Compat.qr(A, Val(false), full=true)
	@test length(f) == 2
	@test size(f[1]) == (3, 3)
	@test f[1] * [f[2]; [0 0]] ≈ A
	f = Compat.qr(A, Val(true), full=false)
	@test f == Compat.qr(A, Val(true))
	@test length(f) == 3
	@test size(f[1]) == (3, 2)
	@test f[1] * f[2] ≈ A[:,f[3]]
	f = Compat.qr(A, Val(true), full=true)
	@test length(f) == 3
	@test size(f[1]) == (3, 3)
	@test f[1] * [f[2]; [0 0]] ≈ A[:,f[3]]
	end

But in JuliaLang/julia#27212 the LinearAlgebra API has been changed, qr doesn't take a full (thin in 0.6) keyword anymore, making these tests fail. Instead, both the "full" and the "thin" representation can be obtained from the returned object.

Ideally, we provide a qr that does the same and deprecate the current method to that. But I have no idea whether that is feasible.

[Sacha0]

I err to agreeing with @fredrikekre in #555 (comment), i.e. simply versioning the tests and moving on. Best!