CLI + output parity: SAM/SJ/read-input knobs and the STAR limit surface

[BenjaminDEMAILLE]

CLI and output parity: 31 STAR 2.7.11b parameters, and a test that makes the parameter surface a machine-checked number.

What changed

tests/parameter_surface.rs checks every user-facing parameter name STAR 2.7.11b accepts against what clap recognises. The name list is the column-1 names of STAR's parametersDefault, minus the three non-user meta-parameters (parametersFiles, sysShell, versionGenome): 200 names, interface only, no STAR source.

Two supporting lists stop "accepted" from meaning "implemented":

• ACCEPTED_BUT_INERT — 22 knobs that parse but change no output byte, each with its reason. A second test asserts the CLI does accept them, so the annotation cannot drift.
• NOT_YET_ACCEPTED — the 26 still missing, grouped by the theme that will bring them.

The test fails in both directions: an unrecognised name that is not declared missing fails, and a name still listed as missing that has since been implemented also fails.

143/200 → 174/200 accepted.

Flags that do something:

Flag	Behaviour
--outSAMmode None	routes to the null writer; SJ.out.tab, Log.final.out and counts are still produced, only records are dropped
--outSAMmode NoQS	strips quality strings
--outSJtype None	suppresses SJ.out.tab
--outSJfilterReads Unique	only unique reads count towards outSJfilter*, so a junction supported solely by multimappers drops out
--outSAMheaderHD / PG / CommentFile	build @hd, extra @pg and @co, with TAG:value validated rather than passed through
--readFilesPrefix	applied during parsing, so every consumer sees final paths
--readQualityScoreBase, --outQSconversionAdd	one signed shift applied as the FASTQ is read, normalising to Phred+33 once
--readNameSeparator	cuts read names at the first separator

Why

Parameter coverage was a claim in prose. A test makes it a number that cannot drift, and the grouped NOT_YET_ACCEPTED list makes the remaining gap legible rather than a bare count.

NoQS is a decorator around the boxed writer rather than a branch at each of the five record-building sites, so the clone is paid only when the flag is set.

One output change at default settings

STAR's default --readNameSeparator is / and STAR cuts there; this codebase did not. On a read named sim_0_YDR305C_mRNA_j567/1, verified against native STAR 2.7.11b:

STAR          sim_0_YDR305C_mRNA_j567
origin/main   sim_0_YDR305C_mRNA_j567/1
this branch   sim_0_YDR305C_mRNA_j567

Verification

Gate green: lib + integration tests, clippy -D warnings, cargo fmt --check, MSRV 1.89. New integration tests cover each implemented flag; the surface test covers the rest.

Bookkeeping note: four flags in NOT_YET_ACCEPTED here (--genomeType, --genomeTransformOutput, --genomeSuffixLengthMax, --sjdbInsertSave) are implemented on #109. Whichever merges second moves them out of the list in the same commit.

🤖 Generated with Claude Code