--soloUMIfiltering MultiGeneUMI_All is aliased to MultiGeneUMI (matches neither STAR nor the docs)

[BenjaminDEMAILLE]

Summary

--soloUMIfiltering MultiGeneUMI_All currently resolves to the same behaviour as MultiGeneUMI:

// src/solo/count.rs:124-125
// MultiGeneUMI_All behaves like MultiGeneUMI for the count matrix.
"MultiGeneUMI" | "MultiGeneUMI_All" => Ok(Self::MultiGeneUmi),

(same alias documented at src/params/mod.rs:1090).

That is neither what STAR 2.7.11b does nor what the option is documented to do, so rustar is currently the only one of the three behaviours that nobody asked for.

What STAR 2.7.11b actually does

In STAR, MultiGeneUMI_All is a no-op. The option is parsed and stored, but the consumption site tests only the MultiGeneUMI flag, so selecting MultiGeneUMI_All alone leaves the multi-gene UMI filter entirely disabled. The counts you get are the unfiltered counts.

What it is documented to do

STAR's documentation describes the two as:

• MultiGeneUMI: remove the lower-count gene's reads when a UMI is assigned to more than one gene.
• MultiGeneUMI_All: when a UMI is assigned to more than one gene, remove it from all of those genes, not just the losers.

So MultiGeneUMI_All is strictly more aggressive than MultiGeneUMI. Aliasing it to MultiGeneUMI silently under-filters; STAR's own no-op silently does not filter at all.

Three options

1. Reproduce STAR's no-op. Maximum byte-fidelity: a run with --soloUMIfiltering MultiGeneUMI_All matches STAR exactly. Also means the flag does nothing, which will surprise anyone who reads the STAR docs.
2. Implement the documented behaviour (split the enum variant, drop cross-gene UMIs from every gene they touch). Diverges from STAR on this one flag, with a documented entry and a locking test.
3. Status quo (alias to MultiGeneUMI). Matches neither.

Resolution

Option 2, implemented in #152:

• a separate UmiFiltering::MultiGeneUmiAll variant rather than an alias;
• an entry in a new docs-old/dev/divergences.md recording the deviation and the reason;
• test_solo_multigene_umi_all_drops_cross_gene_umis asserting the correct result, never STAR's no-op.

Only the MultiGeneUMI_All path changes; the default (-) and the other modes produce identical output. If option 1 is preferred instead, the test asserts the behaviour either way and inverting it is a one-line change.