Benchmark scry using in-memory vs `DelayedArray` objects

[stephaniehicks]

Tasks for @stephaniehicks:

Compare (1) in-memory and (2) DelayedArray objects for increasing sizes of observations and fixed number of features:

Immediate work

• Using Poisson deviance residuals, plot time (y-axis) for increasing observations (x-axis)
  - Track time for both scry::nullResiduals() and BiocSingular::runPCA() separately
  - Try BiocSingular::runPCA() with ExactParam() and IrlbaParam()
• Using Poisson deviance residuals, plot memory-usage (y-axis) for increasing observations (x-axis)
  - Track memory-usage for both scry::nullResiduals() and BiocSingular::runPCA() separately
  - Try BiocSingular::runPCA() with ExactParam() and IrlbaParam()

Near in the future work

@kstreet13 is currently working on implementing the code for the Poisson Pearson and Binomial Pearson cases. Once complete, @stephaniehicks will do the following:

• Using Poisson Pearson residuals, plot time (y-axis) for increasing observations (x-axis)
• Using Poisson Pearson residuals, plot memory-usage (y-axis) for increasing observations (x-axis)
• Using Binomial Pearson residuals, plot time (y-axis) for increasing observations (x-axis)
• Using Binomial Pearson residuals, plot memory-usage (y-axis) for increasing observations (x-axis)

Really in the future work

These last two cases, we will set aside for now.

• Using Binomial deviance residuals, plot time (y-axis) for increasing observations (x-axis)
• Using Binomial deviance residuals, plot memory-usage (y-axis) for increasing observations (x-axis)