solve type stability of DiskIndex

[rafaqz]

Closes #266

I did a bit of code cleanup to understand what was happening, removed all broadcasts over tuples.

But the final thing that worked was solving this :
JuliaLang/julia#48059 (comment)

With really dumb methods like these that somehow trick the compilers recursion detection :

# This makes 3 arg type stable
function _resolve_indices(chunks::Tuple{<:Any}, i::Tuple{<:Any}, indices_pre::DiskIndex, strategy::BatchStrategy)
    inow = first(i)
    indices_new, chunksrem = process_index(inow, chunks, strategy)
    return merge_index(indices_pre, indices_new)
end
# This makes 4 arg type stable
function _resolve_indices(chunks::Tuple{<:Any,<:Any}, i::Tuple{<:Any,<:Any}, indices_pre::DiskIndex, strategy::BatchStrategy)
    inow = first(i)
    indices_new, chunksrem = process_index(inow, chunks, strategy)
    return _resolve_indices(chunksrem, tail(i), merge_index(indices_pre, indices_new), strategy)
end
# ... out to 4 Anys in the Tuples for 6 args of type stability

[lupemba]

Could me merge and release this PR?

[rafaqz]

Can you review it? ;)

[lupemba]

Can you review it? ;)

I did some small test with the PR and had a closer look at the code.

• There are probably a more elegant way to implement the specific _resolve_indices methods but it is hard to get correct. I gave it a try using Val{dimension} for dispatch but I couldn't get it to work.
• I don't understand all possible ways of indexing in DiskArrrays and I am therefore unsure if all possible ways of indexing are inferred now.

My review, this looks good to merge. It fixes the concrete issues with type inference and it doesn't have any negative side effects. There might be some corner cases that are missed but I still think DiskArrays is better with this PR.