Give process_dataset the seed it was already using

[rcannood]

Describe your changes

process_dataset/script.R does this:

cat("Using seed ", par$seed, "\n", sep = "")
set.seed(par$seed)

but --seed is not among the component's arguments -- the Parameters group only declares --dataset_id and --swap. So par$seed is NULL, the log line prints Using seed, and set.seed(NULL) re-seeds from the clock. Both stochastic steps -- the cell-type-balanced downsample to 1000 test cells, and the downsample to 10000 ATAC peaks -- were therefore not reproducible, and there was no way to pin them.

comp_process_datasets.yaml does declare --seed with default: 1, and it's documented in the README, so the knob was advertised but wired to nothing. The workflow's fromState didn't forward it either.

Mirroring the sampling code from script.R:105-110:

seed = NULL (what the component did):
  run A: 5 8 28 41 45
  run B: 1 7 8 22 25
seed = 1 (what it does now):
  run A: 4 5 6 11 37
  run B: 4 5 6 11 37

Verified end to end by running the component twice on bmmc_multiome -- it now logs Using seed 1.

One note on why this hid for so long: neither random branch fires on the test resources. bmmc_multiome has 427 test cells (< 1000) and 1500 peaks (< 10000), so the outputs are byte-identical regardless of seed there. It only matters on the full datasets.

Part of a series of PRs coming out of a pre-run review of the benchmark.

Checklist before requesting a review

• I have performed a self-review of my code

• Check the correct box. Does this PR contain:

  • Breaking changes
  • New functionality
  • Major changes
  • Minor changes
  • Bug fixes

• Proposed changes are described in the CHANGELOG.md

• CI Tests succeed and look good!