Simplify dream contrast handling

[grst]

I was running into an error in the repeated call of eBayes after fit.contrasts.

r$> fit2 <- eBayes(fit2, proportion = opt$proportion,
stdev.coef.lim = stdev_coef_lim_vals,
trend = opt$trend, robust = opt$robust,
winsor.tail.p = winsor_tail_p_vals)
Error in tmp[, j, drop = FALSE] : subscript out of bounds

r$> traceback()
7: [.MArrayLM2(out, , realcoef)
6: out[, realcoef]
5: is.data.frame(x)
4: rowMeans(out[, realcoef]$df.total)
3: .standard_transform(fit_eb, sigma = sqrt(fit_eb$s2.post))
2: eBayes(fit2, proportion = opt$proportion, stdev.coef.lim = stdev_coef_lim_vals,
trend = opt$trend, robust = opt$robust, winsor.tail.p = winsor_tail_p_vals)
1: eBayes(fit2, proportion = opt$proportion, stdev.coef.lim = stdev_coef_lim_vals,
trend = opt$trend, robust = opt$robust, winsor.tail.p = winsor_tail_p_vals)

It turns out that according to the dream vignette the recommended way of setting contrasts it to directly pass it to the dream() call via L=.

Using nobars() + design() before calling dream, we can construct the contrast vector beforehand also avoiding the manipulation of matrix dimnames after the dream call.

In summary, I think this version is cleaner, more computationally efficient, and solves the specific error I had.

PR checklist

Closes #XXX

• This comment contains a description of changes (with reason).
• If you've fixed a bug or added code that should be tested, add tests!
• If you've added a new tool - have you followed the module conventions in the contribution docs
• If necessary, include test data in your PR.
• Remove all TODO statements.
• Broadcast software version numbers to topic: versions - See version_topics
• Follow the naming conventions.
• Follow the parameters requirements.
• Follow the input/output options guidelines.
• Add a resource label
• Use BioConda and BioContainers if possible to fulfil software requirements.
• Ensure that the test works with either Docker / Singularity. Conda CI tests can be quite flaky:
  • For modules:
    • nf-core modules test <MODULE> --profile docker
    • nf-core modules test <MODULE> --profile singularity
    • nf-core modules test <MODULE> --profile conda
  • For subworkflows:
    • nf-core subworkflows test <SUBWORKFLOW> --profile docker
    • nf-core subworkflows test <SUBWORKFLOW> --profile singularity
    • nf-core subworkflows test <SUBWORKFLOW> --profile conda

[grst]

@delfiterradas et al. could you please take that over and make sure the change trickles down into differentialabundance?

[grst]

pre-commit failure is unrelated (API rate limit)

[pinin4fjords]

Backing the direction here from the differentialabundance side: it matches the upstream dream-method.Rd example and removes the double-eBayes flaw. I pulled the branch, applied a tightened version of @delfiterradas's single-coef branch suggestion, and ran the full module nf-test suite under docker against the latest variancePartition container - all 9 tests pass. Inline suggestion below rewrites the contrast block.

The seeded test's dream.results.tsv md5 in tests/main.nf.test.snap shifts under the new single-eBayes flow (numerics still match the existing value assertions in the no-seed tests, so the change is just precision-level reordering). Two occurrences to bump (snapshot file isn't in this PR's diff so I can't suggest it inline):

• 90a06e6b4945c706025582a4c9fddf2c → 865d9c5da7d3749e371231bd2e3727b3

Non-blocking: makeContrastsDream() would be the strictest-correct path but would reject the existing 2 * treatmenthND6 - treatmentmCherry test (weights don't sum to ±1) - worth its own discussion separate from this PR.