feat(18): drop the local utils.py

[tdudgeon]

Second and final PR for this repo under InformaticsMatters/squonk2-jobs#18,
after #24 and #25 which adopted the shared CLI helpers.

Summary

The 35 modules importing the local utils now take it from
dm_job_utilities, which provides all of log, expand_path,
read_delimiter, is_type and calc_geometric_mean. Importing the module
rather than the names — from dm_job_utilities import utils — leaves every
utils.x call site unchanged, so the diff is one line per file.

utils.py (157 lines) is deleted. Net −206/+109.

The two #18 blockers

This repo is the only consumer that actually reached them.

round_to_significant_number → sigfig.round

5 uses, all in dmpk/pk_tmax_cmax_sim.py (Tmax, Cmax, Kel, Ka, V/F at 3 s.f.).

The local implementation leaned on builtin round() and inherited its half-way
behaviour — job-utilities' README calls it inaccurate, and that checks out:

value	local	sigfig
2.675	2.67	2.68
1.005	1.0	1.01

Over 12,000 randomly generated PK-plausible values, none differ. The
divergence needs a value sitting exactly on a decimal boundary, which typed
constants produce and simulation output effectively never does. So this is a
correctness fix with no expected change to real output — but it is a numeric
change to a pharmacokinetics Job, which is why it was kept out of #24.

Wrapped locally as sigfig_round() to pass warn=False: sigfig warns when a
value carries fewer significant figures than requested (e.g. 7.0 at 3 s.f.),
which is common here and was silent before. Without it the Job log would fill
with UserWarning.

get_path_from_digest → digest_utils.py

Used by two scripts, assemble_conformers.py and
prepare_enum_conf_lists.py. job-utilities dropped it as obsolete, but the
sharded directory layout it describes is still what one script writes and the
other reads back — it is not obsolete here. Moved to its own small module rather
than inlined into both. Verified identical to the original across 2,020 cases
(500 real digests × 4 chars/levels configurations, plus degenerate inputs).

Dockerfiles

Every Dockerfile installing job-utilities is bumped to >= 1.4.0.

This was load-bearing, not tidying: dm_job_utilities.utils does not exist
before 1.3.0 — I checked the published wheels. Dockerfile-dmpk and
Dockerfile-rdock were pinned at 1.0.1, and Dockerfile-fns and
Dockerfile-plants at 1.1.1, so all four would have failed at import.

sigfig is added to Dockerfile-dmpk, which did not carry it (Dockerfile-prep
already did, for sa_score.py).

Verification

Every utils.* reference in the repo was resolved against the shared module
programmatically — no unresolved names.

Output parity against main, same inputs:

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

max_min_picker.py differs — but it differs against itself run twice on
main. It's non-deterministic, so unrelated to this change.

assemble_conformers.py and prepare_enum_conf_lists.py (the digest consumers)
import cleanly. dmpk/pk_tmax_cmax_sim.py can't be imported locally
(matplotlib absent — identical on main), so its changed rounding function was
exercised directly.

All 36 touched Python files parse.

jote was not run — needs the affected images rebuilt. Worth a CI run,
particularly for the dmpk and rdock images whose pins moved furthest.

🤖 Generated with Claude Code