feature: Add Nuclei per Cell plot; fix Fraction of Transcripts in Nucleus

[nmalwinka]

Summary

This PR adds a new QC plot to the Xenium report and fixes a latent bug in an
existing plot. Both changes touch nucleus-related metrics and are bundled
together because they share code paths and data sources.

1. New: "Nuclei per Cell" stacked bar

Adds a per-sample stacked bar showing the distribution of cells grouped by
number of segmented nuclei per cell:

• 0 nuclei (red) — cells with no detected nucleus. Useful as an
  under-segmentation / boundary-stain-failure signal.
• 1 nucleus (green) — the expected majority for most tissues.
• 2+ nuclei (blue) — multi-nucleated cells. May reflect merge errors
  during segmentation, or genuine biology (skeletal muscle, cardiac).

Data source: nucleus_count column already present in cells.parquet.
Computed in the existing lazy polars scan in parse_cells_parquet; no
new file I/O. Schema-guarded by if "nucleus_count" in schema:, so older
Xenium outputs that lack the column skip the section gracefully.

Also adds two hidden general-stats columns: % 0-Nuclei Cells (Reds) and
% Multi-Nuclei Cells (Purples). Hidden by default so they don't crowd
the table; users can opt them in via "Configure Columns". Values stored
as fractions in [0, 1]; rendered as percentages via
modify: lambda x: x * 100.0 to match the fraction_transcripts_decoded_q20
convention.

2. Bug fix: "Fraction of Transcripts in Nucleus"

The existing plot computed its metric as:

nucleus_count / total_counts   # both from cells.parquet

This is wrong on as nucleus_count is not a transcript count. Per the Xenium schema it is the count of segmented nuclei per cell (typically 0, 1, or 2+). Dividing it by total_counts produces a meaningless tiny number.

The plot is now correctly derived from transcripts.parquet using the
per-transcript overlaps_nucleus flag, grouped by cell_id (with the
UNASSIGNED sentinel filtered):

fraction_in_nucleus = sum(overlaps_nucleus) / count(*) # per cell_id

The plot function itself (xenium_nucleus_rna_fraction_plot)
is unchanged — only its data source moves from parse_cells_parquet to
parse_transcripts_parquet (one-line edit at the call site).

Bumps version to 1.1.0.

[nmalwinka]

Added Bug fix: single-sample density plots — three cell-related sections ("Cell Area Distribution", "Nucleus to Cell Area", and the transcripts-per-cell side of "Distribution of Transcripts/Genes per Cell") were silently skipping on
single-sample reports because the parser stored only summary statistics, while the density helpers require raw per-cell values. parse_cells_parquet now emits cell_area_values, nucleus_to_cell_area_ratio_values, and
total_counts_values alongside the existing box-stats summaries, so all three sections render as KDE density curves on single-sample reports (matching the existing helptext). Multi-sample reports are unaffected.