add cell type proportion metrics

[RhineCloud]

Changes proposed in this pull request:

• re-worked mean_proportion_deviation function for global metrics to allow 4 different options ["union", "intersection", "sp_specific", "sc_specific"] in one, with improved performance
• removed now superfluous functions for non-common/modality-specific cell types
• added local version of cell type proportions metrics
  • absolute
  • 0: proportions in grid field highly inconsistent with single cell reference
  • 1: proportions in grid field very consistent with single cell reference
  • relative
  • -1: cell types with proportions lower in spatial than single cell data are more common
  • +1: cell types with proportions higher in spatial than single cell data are more common
• added "celltype_proportions_abs" and "celltype_proportions_rel" as options in metrics wrapper function

[LouisK92]

I adjusted the computation for the abs scores, was meant to be very similar as for pos and neg scores.

Also I didn't run into warnings from the line that you in a with statement:

    with pd.option_context("mode.chained_assignment", None):
        df_cells["sp_minus_sc"] = df_cells[obs_key].map(df_props["sp_minus_sc"]).astype(float)

Therefore took the with ... out - what was the reason for it?

[RhineCloud]

I kept getting SettingWithCopyWarnings even after multiple variations of using different indexing methods, including loc as was suggested in the warning, but maybe it worked without setting off the warning after all?

[LouisK92]

I see, yea seems like it somehow works so far at least when I tried.. ^^ Let's keep it like this for now.