simplify array creation when not Offsetjagg

[aminnj]

When there's no 10 byte offset, VectorOfVectors can use the raw offsets almost directly. Basically all that's needed is

offset = convert(Vector{Int64}, rawoffsets)
offset .÷= _size
offset .+= 1

I got a 0-10% speedup on the uncompressed DoubleMu file with this change. If you know of faster replacement for those 3 lines, we could go even faster! 🚀 :)

[aminnj]

@Moelf I should be able to get rid of offset = convert(Vector{Int64}, rawoffsets) and use the Int32 offsets due to JuliaArrays/ArraysOfArrays.jl@eca846a, but I still get complaints, even though my ArraysOfArrays is v"0.5.4"...

[aminnj]

Thanks to @Moelf for the magic lines to get Int32 offsets everywhere. If there is any speedup, it's slight, but not negative at least.

[Moelf]

https://github.com/tamasgal/UnROOT.jl/blob/65bc44d4af4cdc80a94fe7479017d6d46fa5e80b/src/custom.jl#L56

you also need to change this, it's alarming our test doesn't detect this

[aminnj]

Test added separately in #92

[aminnj]

Since

julia> Tuple{SubArray{UInt8, 1, Vector{UInt8}, Tuple{UnitRange{Int64}}, true}, Vector{Int32}} <: Tuple{Vector{UInt8},Vector{Int32}}
false

that means we were incurring the non-trivial cost of convert() at the end of readbasketseek(). However, we can't just remove the type annotation because then there are downstream issues with the @view rawdata, so the @view is removed. Total allocations slightly decreases, but performance is basically the same.