`vst()` fails when using custom latent variable with `vst.flavor = "v2"`

[blasks]

When calling sctransform::vst() with a user-defined latent variable and vst.flavor = "v2", the function errors with:

Error in fit_glmGamPoi_offset(...): !is.null(log10_umi) is not TRUE

This happens even when a the column in cell_attr is a valid numeric column.

Minimal Example

library(sctransform)
library(Matrix)

# fake count data
set.seed(1)
data <- Matrix(
  rpois(50, lambda = 5), 
  nrow = 10, 
  ncol = 5,
  dimnames = list(paste0("gene", 1:10), paste0("cell", 1:5)), 
  sparse = TRUE
)

# fake cell attributes
cell_attr <- data.frame(
  user_var = runif(5, 100, 1000),
  row.names = colnames(data)
)

# works fine
result_ok <- vst(data, cell_attr = cell_attr, vst.flavor = "v2", verbosity = 2)

# fails
result_fail <- vst(data, cell_attr = cell_attr, latent_var = "user_var", vst.flavor = "v2", verbosity = 2)

# works, but with a different result from the first call
cell_attr$log_umi <- log10(cell_attr$user_var)
result_different <- vst(data, cell_attr = cell_attr, latent_var = "log_umi", vst.flavor = "v2", verbosity = 2)

I'm running sctransform version 0.4.2, and R version 4.5.1 on a macOS system.

Thanks for looking into this!