feat: add geneBody_coverage and read_GC RSeQC tools

[BenjaminDEMAILLE]

Closes #127.

Adds the two remaining RSeQC tools, both inside the existing single pass.

geneBody_coverage

src/rna/rseqc/gene_body_coverage.rs — 5' → 3' coverage in 100 percentile bins.

• Each transcript's exonic bases are reduced to 100 percentile positions with a direct port of RSeQC's mystat.percentile_list
• Reverse-strand transcripts are flipped so bin 1 is always the 5' end
• Transcripts under 100 bp are skipped (RSeQC's --minimum_length default)
• Read filters: unmapped, QC-fail, secondary, duplicate; deletions and N skips do not count as covered

Outputs {sample}.geneBodyCoverage.txt (RSeQC format), .geneBodyCoverage.r, and a native .curves.png/.svg.

One deliberate deviation, documented in the docs and module: RSeQC profiles every entry of a BED12 reference (usually a housekeeping-gene set); RustQC has no BED input here and profiles one representative transcript per gene from the GTF — the one with the most exonic bases, the same choice the TIN analysis already makes.

⚠️ Worth a maintainer's eye: the port needs Python's round-half-to-even. With f64::round (half away from zero), interpolated percentile positions land one base off and 7 of 100 bins came out ±1 against upstream. There's a unit test pinning the rounding behaviour and a note in AGENTS.md.

read_GC

src/rna/rseqc/read_gc.rs — GC distribution of mapped reads.

• Filters match upstream ParseBAM.readGC: skip unmapped, QC-fail and below the MAPQ cutoff; secondary alignments and duplicates are counted (upstream does not filter them)
• GC percentage formatted to two decimals, as upstream does

Outputs {sample}.GC.xls, .GC_plot.r, and a native .GC_plot.png/.svg.

Two deliberate deviations: rows are sorted ascending by GC percentage (upstream emits Python dict order, which is BAM-order dependent and not reproducible under parallel processing), and reads stored without SEQ are skipped with a warning instead of dividing by zero.

Validation against RSeQC 5.05

Ran upstream geneBody_coverage.py and read_GC.py (RSeQC 5.05, pysam) on tests/data/test.bam, using a BED12 built from tests/data/test.gtf with the same representative-transcript choice:

• gene body coverage: all 100 bins identical
• GC table: identical (0.00 483)

Both references are pinned in new integration tests. Also verified --threads 4 produces byte-identical output to single-threaded, and that the skip flags remove the outputs.

Other

• Config: rna.gene_body_coverage.enabled, rna.read_gc.enabled (both default true); CLI --skip-gene-body-coverage, --skip-read-gc
• Docs: new sections on the RSeQC page (including a note on how this differs from the Qualimap gene body coverage RustQC already produces), config/CLI reference entries, "8 RSeQC tools" → 10 where claimed
• cargo test — 211 lib + 20 integration tests pass; cargo fmt --check, cargo clippy -- -D warnings clean

🤖 Generated with Claude Code