The second of three planned PRs for LSDB's #449.

Overall plan:

• First PR: change the directory names to mem_size_histogram and row_count_histogram
  • Rename histogram directory (histograms/ -> row_count_histograms/) #602
• Second PR: create the mem_size_histogram (but don't do anything with it)
  • 📍You are here
• Third PR: use the mem_size_histogram to do the partitioning

A little more detail:

1. Set up stage:
   1. We initialize the ResumePlan to have a concept of threshold_mode. The threshold mode defaults to row_count, but is set to mem_size if byte_pixel_threshold is in the input args (and is not None).
   2. We also specify some paths in ResumePlan: MEM_SIZE_HISTOGRAM_BINARY_FILE and MEM_SIZE_HISTOGRAMS_DIR.
   3. We run ResumePlan's gather_plan, which creates histogram directory(/ies) among other set up stuff.
2. Mapping stage:
   1. Here's where we map input files to Healpix pixels (via the call to map_reduce's map_to_pixels); and in doing so, we create the histogram. I am electing to make two histograms here, so long as the histogram mode is set to mem_size. Is is ok to just call mr.map_to_pixels a second time like that?
   2. We add memory size calculating method _get_mem_size_of_chunk and its two helpers _get_row_mem_size_data_frame and _get_row_mem_size_pa_table
3. Binning stage:
   1. No changes for now, except that we add an explicit parameter which_histogram to read_histogram to show that we're reading the row_count histogram. It's the default, but I wanted to include it for readability/safety.