avoid allocation when negating BitArray

[OkonSamuel]

avoid allocation when negating BitArray in dropmissing! method

## this pr
julia> Random.seed!(0);

julia> rv1 = rand(Bool, 10000);

julia> rv2 = rand(Bool, 10000);

julia> mv1 = [rv1[i] ? missing : i for i in eachindex(rv1)];

julia> mv2 = [rv2[i] ? missing : Float64(i) for i in eachindex(rv2)];

julia> df1 = DataFrame(i = 1:10000, x=mv1, y=mv2);

julia> df2 = DataFrame(i = 1:10000, x=mv1, y=mv2);

julia> df3 = DataFrame(i = 1:10000, x=mv1, y=mv2);

julia> @btime dropmissing!($df1, :x);
  4.697 μs (18 allocations: 5.28 KiB)

julia> @btime dropmissing!($df2, :y);
  5.310 μs (18 allocations: 5.28 KiB)

julia> xy = [:x, :y];

julia> @btime dropmissing!($df3, $xy);
  18.728 μs (46 allocations: 11.34 KiB)

## master
julia> Random.seed!(0);

julia> rv1 = rand(Bool, 10000);

julia> rv2 = rand(Bool, 10000);

julia> mv1 = [rv1[i] ? missing : i for i in eachindex(rv1)];

julia> mv2 = [rv2[i] ? missing : Float64(i) for i in eachindex(rv2)];

julia> df1 = DataFrame(i = 1:10000, x=mv1, y=mv2);
df2
julia> df2 = DataFrame(i = 1:10000, x=mv1, y=mv2);

julia> df3 = DataFrame(i = 1:10000, x=mv1, y=mv2);

julia> @btime dropmissing!($df1, :x);
  5.281 μs (21 allocations: 6.06 KiB)

julia> @btime dropmissing!($df2, :y);
  5.766 μs (21 allocations: 6.06 KiB)

julia> xy = [:x, :y];

julia> @btime dropmissing!($df3, $xy);
  19.032 μs (48 allocations: 11.77 KiB)

[bkamins]

Looks good. Can you submit here some performance benchmarks for comparison before we merge? Thank you!

[OkonSamuel]

Looks good. Can you submit here some performance benchmarks for comparison before we merge? Thank you!

sure.

[OkonSamuel]

@bkamins done

[bkamins]

Thank you!