Fix type piracy issue

[tlnagy]

Fixes the issue brought up by @timholy in #36 (comment). Added this branch to testing as well.

[tlnagy]

This also fixes a pretty serious bug that caused TiffImages to ignore stripped offsets, which actually yielded a substantial performance improvement. So this PR introduces a ~5x performance regression vs master. This condition is likely common enough that it's worth having a separate branch to handle the case were an image is striped but the stripes are contiguous and can therefore be read in as a single block.

[tlnagy]

Checking for contiguous strips allows me to recover the performance of master.

julia> using TiffImages

julia> @time TiffImages.load("/home/tlnagy/Downloads/Gcamp6F_BSA_60px_3_stack.ome.tif");
Progress: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| Time: 0:00:01
  2.724760 seconds (4.75 M allocations: 738.599 MiB, 2.18% gc time)

julia> @time TiffImages.load("/home/tlnagy/Downloads/Gcamp6F_BSA_60px_3_stack.ome.tif");
  0.768867 seconds (286.85 k allocations: 513.183 MiB, 3.43% gc time)

julia> using ImageMagick

julia> @time ImageMagick.load("/home/tlnagy/Downloads/Gcamp6F_BSA_60px_3_stack.ome.tif");
  4.143202 seconds (2.69 M allocations: 1.540 GiB, 5.18% gc time)

julia> @time ImageMagick.load("/home/tlnagy/Downloads/Gcamp6F_BSA_60px_3_stack.ome.tif");
  3.236719 seconds (6.72 k allocations: 1.408 GiB, 0.72% gc time)

[tlnagy]

I created an artificial image that has its stripes offset by a random amount that should catch any regressions in handling discontiguously striped uncompressed images.