Deprecate v0 in accumulate in favor of init keyword

[andyferris]

Implements the same as #27711 for accumulate and accumulate!.

[Keno]

Looks like we have a couple of test errors on this PR.

[andyferris]

I got some time to look at this again. Funniliy enough, I had to add functionality to make this work cleanly. 😆

With this change, init should work on arrays of arbitrary dimension, not just vectors. Unfortunately, I haven't the time today to wrap this up (and provide tests for the new functionality).

[andyferris]

For some reason this might not be working with offset arrays... I'm actually not sure how I broke that?

[andyferris]

Also @StefanKarpinski not sure if this deserves a similar milestone to the reduce PR? (Just so we don't forget to merge this).

[andyferris]

So I found some questionable behavior regarding indices of OffsetArrays.

julia> include("test/TestHelpers.jl")
Main.TestHelpers

julia> using Main.TestHelpers.OAs

julia> a = [11,12,13]
3-element Array{Int64,1}:
 11
 12
 13

julia> b = OffsetArray(a, (-3,))
OffsetArray{Int64,1,Array{Int64,1}} with indices -2:0:
 11
 12
 13

julia> i = LinearIndices(b)
LinearIndices{1,Tuple{Base.Slice{UnitRange{Int64}}}} with indices -2:0:
 -2
 -1
  0

julia> i[-2]
-2

julia> i[-2:end]
-5:-3

I suspect this is a bug (cc @timholy @mbauman). Because these are a Slice I feel the scalar getindex behavior is correct and the behavior here is wrongly assuming 1-based indexing. The second commit of the PR attempts (and fails now succeeds...) to work around the inconsistency between scalar and range getindex on these indices by invoking iteration directly.

[andyferris]

xref #27986

[andyferris]

Pending further comments, if this passes CI I'll merge (only 4 milestone items left, and this is one of them!)

[mbauman]

Travis failures were FileWatching for Mac and in dep-gathering for Linux.

[fredrikekre]

Cheeky little comma that breaks CI!