feat(18): adopt the shared CLI helpers in the rdkit Jobs

[tdudgeon]

Sixth consumer migration for InformaticsMatters/squonk2-jobs#18, and the first
of two PRs for this repo.

Summary

Replaces the hand-typed "Input/output options" block in seven scripts with
rdkit_utils.add_common_molecule_io_args(), and the interval reporting idiom
with dm_job_utilities' add_reporting_args() / ProgressReporter:

rdkit_props.py, rdkit_dedup.py, sa_score.py, cluster_butina.py,
enumerate.py, le_conformers.py, im_mordred/descriptor_generator.py

Net −109/+45 lines. Dockerfile-prep and Dockerfile-mordred bumped to
im-data-manager-job-utilities>=1.4.0 / im-rdkit-utilities>=1.1.2.

No manifest changes. --input and --output are retained as aliases by the
helper, so every existing command block keeps working — this is exactly what the
alias design in
squonk2-jobs#17
was for. The sweep to canonical --infile/--outfile is deferred to before the
release that drops the aliases.

⚠️ screen.py is excluded — a real collision

It uses -d for --descriptor, which collides with the helper's -d for
--delimiter:

argparse.ArgumentError: argument -d/--delimiter: conflicting option string: -d

screen.py is otherwise a strong adopter (9/11 options), so this is a genuine
loss. Resolving it needs a decision:

1. Drop -d from --descriptor. The manifest passes --descriptor
   long-form, so no manifest change — but it removes a public short flag.
2. Give the helper a way to omit -d, e.g. a short_flags knob. Costs an
   im-rdkit-utilities release and adds API surface for one caller.

Left out rather than decided unilaterally.

Other judgement calls

• cluster_butina.py takes the I/O group but not add_reporting_args() —
  it has no progress loop and never had --interval, so adding one would be
  noise.
• sa_score.py, le_conformers.py and enumerate.py gain --omit-fields
  and/or --write-header they did not previously have, as members of the
  shared group. Unused and optional; the accepted cost of a common group.
• --mol-column now defaults to None so SmilesReader infers it. With
  --id-column 0 the old hardcoded 0 pointed the molecule and ID columns at
  the same field. No manifest passes --mol-column, so the default path is
  unchanged.

Verification

Output parity against main, same inputs and arguments:

Script	Result
rdkit_props.py	byte-identical
rdkit_dedup.py	byte-identical
sa_score.py	byte-identical
cluster_butina.py	byte-identical
enumerate.py	byte-identical
le_conformers.py	byte-identical

im_mordred/descriptor_generator.py could not be run locally (mordred not
installed — it fails identically on main). Its parsed namespace is
identical old vs new, and the same file lineage was validated end-to-end
with jote in
squonk2-desc-mordred#3.

Parsed-namespace comparison, old vs new, for all seven. The only differences
are the four intended ones:

• id_column: '1' → 1 (the str_or_int type)
• mol_column: 0 → None (reader inference, above)
• outfile → output (dest rename; call sites updated)
• new omit_fields / write_header keys where the group adds them

Building every parser also proves there are no other short-flag collisions —
screen.py was the only one.

jote was not run — these Jobs need the informaticsmatters/vs-prep image
rebuilt. Worth a CI run before merge.

Follow-up PR

utils.py is untouched here. Removing it needs the two functions #18 flags as
blockers resolved, and this repo is the only consumer that actually hits them:

• round_to_significant_number — 5 uses, all in dmpk/pk_tmax_cmax_sim.py.
  Replacing it with sigfig.round changes numeric output, so it needs its
  own validation and shouldn't ride along with a CLI refactor.
• get_path_from_digest — 1 use in assemble_conformers.py.

🤖 Generated with Claude Code