Order scdesign2 input by spatial cluster

[rcannood]

simulate_count_scDesign2() writes its output one cell type at a time, in contiguous blocks (new_count[, llim:ulim] <- ... in the package source), but the script attached coordinates from the input in its original order. The datasets are not sorted by spatial_cluster -- MOBNEW starts 1,2,3,3,4,2,4,1 -- so every spot received the counts of an unrelated spot, and the spatial signal was destroyed before the metrics saw it.

It shows in the results: scdesign2 came last on ks_statistic_morans_I (25.7, against 1.8 for scdesign3_nb) and last on svg_precision (0.442). Every other method already does input[order(input$obs$spatial_cluster)] first; this one now does too.

While in there: cell_type_sel and cell_type_prop are matched to each other by position, but were built with unique() and table() respectively. Those only coincide when the labels happen to sort the same way -- with 10 or more clusters table() gives 1,10,11,12,2,... and the proportions get attached to the wrong clusters. Both now derive from unique() on the sorted data, which is also exactly the block order:

cell_types       : 1,2,3,4,5,6,7,8,9,10,11,12
blocks in data   : 1,2,3,4,5,6,7,8,9,10,11,12
order matches    : TRUE
block sizes exact: TRUE
sort(unique())   : 1,10,11,12,2,3,4,5,6,7,8,9   <- what it would have been

Block sizes come out exact, so coordinates line up spot for spot rather than approximately.

Checklist before requesting a review

• I have performed a self-review of my code

• Check the correct box. Does this PR contain:

  • Breaking changes
  • New functionality
  • Major changes
  • Minor changes
  • Bug fixes

• Proposed changes are described in the CHANGELOG.md

• CI Tests succeed and look good!