Make isapprox error unless shapes match

[jariji]

I don't like this and I would like it to error

julia> [1; 2; 3;;]
3×1 Matrix{Int64}:
 1
 2
 3
julia> [1; 2; 3;;] ≈ [1,2,3]
true

just like this does

julia> (x for x in 1:3) ≈ [1,2,3]
ERROR: MethodError: no method matching isapprox(::Base.Generator{UnitRange{Int64}, typeof(identity)}, ::Vector{Int64})

[nsajko]

Possible course of action: make a PR then request a PkgEval from NanoSoldier to see what will break.

[nsajko]

I would like it to error

Given that == returns false for array arguments of mismatched shape, perhaps [1; 2; 3;;] ≈ [1,2,3] should return false, too?

[jariji]

I think Matrix == Vector should error too. #40717

[nsajko]

Yeah, but that's definitely for a breaking release.

[nsajko]

Offending method:

julia/stdlib/LinearAlgebra/src/generic.jl

Lines 1877 to 1892 in e516e4c

	# isapprox: approximate equality of arrays [like isapprox(Number,Number)]
	# Supports nested arrays; e.g., for `a = [[1,2, [3,4]], 5.0, [6im, [7.0, 8.0]]]`
	# `a ≈ a` is `true`.
	function isapprox(x::AbstractArray, y::AbstractArray;
	atol::Real=0,
	rtol::Real=Base.rtoldefault(promote_leaf_eltypes(x),promote_leaf_eltypes(y),atol),
	nans::Bool=false, norm::Function=norm)
	d = norm(x - y)
	if isfinite(d)
	return iszero(rtol) ? d <= atol : d <= max(atol, rtol*max(norm(x), norm(y)))
	else
	# Fall back to a component-wise approximate comparison
	# (mapreduce instead of all for greater generality [#44893])
	return mapreduce((a, b) -> isapprox(a, b; rtol=rtol, atol=atol, nans=nans), &, x, y)
	end
	end