deprecate zero-arg Vector constructor

[fredrikekre]

Ref #14201, #20330 (specifically #20330 (comment)) and #20402 (comment)

[nalimilan]

I'm not sure I like this idea. Yes, that would be more consistent with Matrix{T}() throwing an error, but Vector{T}() is quite natural and very common...

[mauro3]

I think the most natural for Vector{T}() is T[], so making this consistent will at least not impact my code. In fact, maybe this PR could use the T[] form where appropriate?

[fredrikekre]

I'm not sure I like this idea. Yes, that would be more consistent with Matrix{T}() throwing an error, but Vector{T}() is quite natural and very common...

I like the consistency, but on the other hand, an empty vector is useful which a empty matrix really isn't, so that could warrant special treatment of Vector()...

I think the most natural for Vector{T}() is T[], so making this consistent will at least not impact my code. In fact, maybe this PR could use the T[] form where appropriate?

I included the [] / T[] syntax as a suggestion in the deprecation warning, I also think [] / T[] is nicer than Vector() / Vector{T}().

[rfourquet]

Another consistency would be broken with this PR when we see Vector as the general go-to container (the list of python, std::vector of C++ etc.), in the same family as other general containers like Set, rather than as a member of multi-dim arrays; then it's natural that a call to the type with no argument constructs an empty container. Consider e.g.

T = rand(Bool) ? Vector{Int} : Set{Int}
c = T()

Similarly, Vector() should not be deprecated IMHO, by analogy to Set().

[fredrikekre]

This is a bad idea, I kinda like Vector{T}() to make a new empty vector, and people seem to agree with me, so I'll close this, and also #21082 and put this issue at a rest.