Handle all-zero features in density calculation

[foertsch]

Fixes #20

This PR makes plot_density() handle all-zero feature weights safely.

Summary

• guard calculate_density() against sum(w) <= 0
• return zero densities instead of propagating invalid values
• add regression tests for all-zero inputs

Why

When the selected feature has all-zero weights, the density calculation currently divides by sum(w), which becomes zero and produces invalid values downstream.

This can cause plot_density() to fail for zero-expression features instead of returning a valid empty/zero-density plot.

What changed

• if sum(w) <= 0 and map = TRUE, return a zero vector of length nrow(x)
• if sum(w) <= 0 and map = FALSE, return a zero-valued density object compatible with the existing downstream code
• add tests covering both calculate_density() and plot_density() on an all-zero feature

Verification

Tested locally with:

testthat::test_file("tests/testthat/test-zero-density.R")
testthat::test_file("tests/testthat/test-plot_density.R")

Also verified directly with a synthetic all-zero metadata feature on SeuratObject::pbmc_small, where plot_density() now returns a ggplot.

Related issue

This likely also addresses issue #29 when the reported Empty raster error is caused by a zero-valued feature, but I have not reproduced the exact original object from that report.