Changes in v0.0.3

Updated support for v4 vs. v5 Seurat objects.
Exposed choice of variational inference algorithm to the user via argument VI.algorithm in findvariableFeaturesBayes().
Updated documentation thoroughly.
Removed QR decomposition on fixed effects as it wasn't necessary.

Changes in v0.0.2

Added conditional support for legacy Seurat v4 objects in addition to the default Seurat v5 objects.
Added per-gene estimated posterior variances and dispersions based on NB variance definition, along with credible intervals for each per-gene.
Implemented ability to select HVGs by either estimated dispersion or estimated variance in classifyHVGs().
Changed model fitting process to support within-chain parallelism as long as enough cores are available.
Added function computeNaiveGeneStatistics() to estimate (in a Frequentist manner) per-gene mean, variance, and dispersion.
Sped up findvariableFeaturesBayes() via:
- Adding GPU acculeration support for OpenCL-compatible devices.
- Performing QR decomposition on covariates before fitting.
- Added compiler optimization flags for Stan to C++ code conversion.
- Set normalize = FALSE in call to brms::brm(), which increases efficiency of sampling from approximate posterior.

Initial package skeleton.
Added main function findVariableFeaturesBayes() (still in development right now).
Added helper function sampleMarginal() to make sampling from posterior marginal distribution possible.
Added support for adding gene statistics metadata to SingleCellExperiment or Seurat objects after estimation.
Changed main model backend to brms via cmdstanr instead of INLA, since INLA doesn't appear to support per-group estimation of the Negative-binomial overdispersion parameter.
Added function classifyHVGs() to add a label to HVGs in Seurat or SingleCellExperiment object metadata based on several different methods.
Added function theme_bayesHVG(), which implements a publication-ready theme for ggplot2.