WIP: refgenie1 migration (do not merge)

[nsheff]

Do not merge. This PR exists so reviewers can read the diff and the
findings; it is not a merge candidate.

Summary

Hard-cut migration of PEPATAC from legacy refgenconf to refgenie 1.0
(refgenie1, the SQLModel-backed reimplementation). Removes the dead
refgenconf import in pipelines/pepatac.py, replaces the populator
hook in sample_pipeline_interface.yaml with refgenie1's
looper_refgenie_populate_local, swaps \$REFGENIE →
\$REFGENIE_DB_CONFIG_PATH in var_templates, renames
bowtie2_index.dir → bowtie2_index.bowtie2_index in Jinja
(refgenie1's bowtie2_index asset class has no dir seek_key —
the prefix path is what bowtie2 -x consumes anyway), updates
requirements.txt and requirements-conda.yml, and rewrites the
refgenie-touching docs to use refgenie 1.0 CLI syntax.

Companion changes (refgenie1)

The looper-style local populator is added on
refgenie/refgenie1@nsheff-refactor-2:

• refgenie/populator.py — looper_refgenie_populate_local (drop-in
  replacement for refgenconf.looper_refgenie_populate)
• refgenie/__init__.py — export
• tests/test_populator.py — unit tests against the session refgenie
  fixture

Two bugs surfaced and were fixed there:

• Refgenie(database_config_path=...) rejected str (Path-only contract,
  no coercion) → coerce to Path in the populator.
• Looper's _update_namespaces requires the populated namespace key
  to pre-exist on input → mutate input dict before returning.

Findings (dogfooding)

The full set of refgenie1 API gaps, naming divergences, asset-class
gaps, CLI/install footguns, and cluster-integration issues found
during this migration are recorded in findings.md at the repo root.
Highlights:

• No Refgenie.list_seek_keys_values() equivalent — the populator
  has to walk every (genome, group, asset, seek_key) leaf manually.
• bowtie2_index.dir is gone (no built-in dir seek_key in refgenie1).
• Refgenie.asset.seek returns Path (refgenconf returned str).
• PyPI name collision: legacy refgenie 0.12.x and refgenie1 1.0.x
  both publish under refgenie. Pin refgenie>=1.0.0.
• \$REFGENIE (legacy) → \$REFGENIE_DB_CONFIG_PATH (refgenie1).
  Anyone with \$REFGENIE in .bashrc needs migration guidance.
• refgenie pull g/a is gone; replaced by refgenie genome init +
  refgenie add g/a --recipe r.
• Refgenie1 has no --version flag and no asset subcommand.
• The cluster's broken legacy-refgenie binary at ~/.local/bin/refgenie
  shadows refgenie1 unless refgenie1.env is sourced.

Validation

End-to-end on Rivanna (SLURM 12499800, 2:16 wall-clock, 4 cores). The
populator delivers a fully-resolved command line; PEPATAC consumes
refgenie1 paths for hg38 (fasta, chrom_sizes, refgene_tss, blacklist,
feat_annotation, bowtie2_index) and rCRSd (fasta, bowtie2_index)
without modification. Pipeline ran through trimming → fastqc →
prealignment → primary alignment → sort/index → dedup → fragment
classification, and FAILED at signal generation in gtars-uniwig due
to a Rust panic on BAM header parsing (unrelated to refgenie1 — a
gtars-rs / PEPATAC bug).

Outputs (sort.bam, dedup BAM, fragment classes) live on cluster at
`/project/shefflab/brickyard/results_analysis/atacbase/forge/pilot/refgenie1_validation/results_pipeline/results_pipeline/test1/`.

Cluster run log:
`/project/shefflab/brickyard/results_analysis/atacbase/forge/pilot/refgenie1_validation/run_validation.log`.

Plan

`assistant/pepatac_refgenie1_branch_plan_v1.md` — Phase 3 of
`assistant/accbase_refgenie1_dogfood_metaplan_v1.md`.

Test plan

• dry-run looper produces fully-resolved sample command
• populator unit tests pass (refgenie1 `tests/test_populator.py`)
• end-to-end SLURM run reaches alignment + dedup
• full pipeline completion (blocked on gtars-rs BAM header fix)
• master-branch parity diff (blocked on full completion)