Add RPKM calculation step; harden RPKM script against zero denominators

[mtinti]

Motivation

• Add a pipeline step to compute RPKM values from featureCounts output so normalized expression matrices are produced alongside count matrices.
• Ensure RPKM outputs are treated as required outputs and are copied to the results directory and considered when checking whether a sample/run-tag is completed.
• Make the RPKM calculation robust to edge cases such as zero total reads or zero feature lengths to avoid divide-by-zero warnings.
• Support featureCounts files processed one-sample-at-a-time (single counting column) without pipeline changes.

Description

• Added Snakemake rules in rules/counting/rpkm.smk implementing rpkm_* rules for paired and single workflows and completion flags, and included the module in Snakefile via the conditional includes.
• Added a dedicated lightweight conda environment envs/rpkm.yaml for the RPKM step and wired it into the rules using conda blocks.
• Implemented rules/scripts/calc_rpkm.py which detects sample columns to the right of Length, computes RPKM for one or more sample columns, logs to Snakemake logs, and writes a tab-delimited RPKM file; the computation now uses numpy.divide(..., where=denominator != 0) to safely handle zero denominators.
• Updated pipeline wiring: added get_rpkm_flag in rules/common.smk, required rpkm_complete in copy_results in rules/results.smk, copy RPKM files to results, and included RPKM checks in rules/sample_utils.py; also updated README.md and the workflow_steps list in Snakefile.

Testing

• No automated tests were executed for these changes.

---

Codex Task