Mount sidecar inputs without placing them on the command line

[jorgeMFS]

Merge order. Stacked on #91 and targets honour-input-filename, not main — merge #91 first, then this. It is also the frontend half of a two-repo change: ieeta-pt/biochef-hub#26 adds the mode to the schema and is itself stacked on ieeta-pt/biochef-hub#25. A recipe cannot use mode: sidecar until both repos have landed, so neither half is useful alone, but neither breaks anything if it lands first.

Full order: biochef-hub#25 → biochef-hub#26, and Biochef#91 → this. The two chains are independent of each other.

Summary

Some tools locate a companion file by name rather than taking it as an argument. htslib resolves an index by string-appending .csi or .bai to the data file path, so bcftools merge, isec and consensus need a file at exactly <datafile>.csi that is never passed to the tool.

Today that is inexpressible. A mode: file input is always emitted — flagged, or positional if it has no flag — so declaring the index as an extra input hands it to the tool as a stray positional and the command fails. This is the gap noted in #91, which said the filename change alone would not unblock those operations.

This treats mode: sidecar exactly like a file input for mounting, and leaves the argument-building block untouched so a sidecar falls through it.

Why the naming works out

The sidecar goes through the same resolver as every other input, so within one invocation:

Declared	Mounted as
data input, filename: a.vcf.gz	input-<nodeId>-a.vcf.gz
sidecar, filename: a.vcf.gz.csi	input-<nodeId>-a.vcf.gz.csi

The data file is passed to the tool as input-<nodeId>-a.vcf.gz; htslib appends .csi and finds the sidecar. The per-invocation prefix is on both, so the sibling relationship survives and names stay unique between nodes.

The upstream-node path is covered too: a sidecar piped from, say, a samtools index node is re-mounted under its declared name by the same code #91 added.

Scope

+8 −2, all within the input-preparation loop. The argument-building block is deliberately not touched — that is what makes a sidecar invisible to argv.

Verification

• Inert until adopted — no recipe declares mode: sidecar on any ref of biochef-recipes (it could not, until the hub change lands), so the new path is unreachable and nothing existing changes behaviour.
• Build — npm run build: webpack 5.106.2 compiled successfully.

No automated tests: the repo has no test script and no workflow runs on a PR here. As with #91, the substantive evidence is that the changed path cannot execute for any recipe that exists today.

---

Merge order

Stacked on #91 — base it on honour-input-filename, not main, and merge #91 first.

src/utils/toolUtils.js is where five open branches meet: #91, #92, #93, this one, and WildBunnie's aioli-vendored (#90). They all conflict there pairwise, so the order matters more than usual.