Avoid Union{Missing,T} in columns after DropMissing

[eliascarv]

This PR adds a post-processing in the apply function of the DropMissing transformation and a pre-processing in the revert function.
This is necessary so that the type of columns where the DropMissing transformation will be applied is T instead of Union{Missing,T} .

[juliohm]

I think this current implementation has the issue that it forces a Union{Missing,T} in the revert even when the original table doesn't have missing values. As we discussed over Zulip, we just need to save the original column eltype during the apply and reuse it in the revert.

[juliohm]

Need to fix the revert so that it restores the original eltype, and also add tests.

[codecov-commenter]

Codecov Report

Merging #35 (3634072) into master (cb72d48) will increase coverage by 0.06%.
The diff coverage is 96.00%.

@@            Coverage Diff             @@
##           master      #35      +/-   ##
==========================================
+ Coverage   92.51%   92.57%   +0.06%     
==========================================
  Files          16       16              
  Lines         414      431      +17     
==========================================
+ Hits          383      399      +16     
- Misses         31       32       +1     

Impacted Files	Coverage Δ
src/transforms/filter.jl	98.00% <96.00%> (-2.00%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cb72d48...3634072. Read the comment docs.

[eliascarv]

Done. I made the requested changes and added tests @juliohm.

[eliascarv]

@juliohm, I moved post and pre processing to apply and revert. This avoids creating the helpers functions.
The _nonmissing function has these two extra methods to avoid unnecessary calls to the collect function:

_nonmissing(::Type{T}, c) where {T} = c
_nonmissing(::Type{Union{Missing,T}}, c) where {T} =
  collect(T, c)

function _nonmissing(columns, col)
  c = Tables.getcolumn(columns, col)
  _nonmissing(eltype(c), c)
end

But I can simplify this function, but the calls to the collect function will happen in all columns of the table:

function _nonmissing(columns, col)
  c = Tables.getcolumn(columns, col)
  collect(nonmissingtype(eltype(c)), c)
end

Which of the two options do you prefer?

[juliohm]

Just a few minor changes now and I think we are ready

[eliascarv]

This PR is ready for a final review @juliohm!