PEtab import: steady-state (time=inf) measurements are materialized as a TimeCourse and crash at config load (OverflowError)

[wshlavacek]

Summary

A PEtab problem whose measurements are at time = inf (steady-state) imports fine, but config load then crashes because the importer materializes the measurement as a TimeCourse rather than a steady-state scan. PyBNF already has steady-state support (a ParamScan with steady_state=>1, ADR-0046); the gap is only that import_job does not select it for time = inf measurements.

Reproduce

Blasi_CellSystems2016 from the Benchmark-Models-PEtab collection (subset I of the Grein et al. 2026 optimizer benchmark). Every Blasi measurement is at time = inf.

from pybnf.petab import petab1to2_preserve_scale, import_job
import tempfile
v2 = petab1to2_preserve_scale("Blasi_CellSystems2016/Blasi_CellSystems2016.yaml", tempfile.mkdtemp())
import_job(v2, "out", job_type="gntr")   # OK — writes an .exp whose time column is `inf`

Then load the emitted conf (or run pybnf -c imported.conf -o):

File ".../pybnf/config.py", line 1422, in _load_experiments
    action = TimeCourse({'suffix': name, 'method': method}, explicit_points=points)
File ".../pybnf/pset.py", line 1792, in __init__
    self.stepnumber = int(np.round((self.time - self.t_start) / self.step))
OverflowError: cannot convert float infinity to integer

Root cause

PEtab time = inf denotes a steady-state measurement. import_job writes the literal inf into the .exp time column and emits a time-course experiment (experiment: ..., method: ode). TimeCourse.__init__ then derives a step count from the (assumed finite) time grid, which overflows on inf.

Expected

Recognize time = inf measurements at import and emit a steady-state action instead — a ParamScan with steady_state=>1, ss_method=>"newton" (ADR-0046) — so the datum is compared at the KINSOL steady state rather than at a fixed endpoint. PyBNF already supports this scan; only the importer's action selection needs to branch on time = inf.

Impact

Blocks Blasi_CellSystems2016 (all measurements at t = inf) — the last open subset-I problem in the Grein et al. 2026 benchmark import. Distinct from #509 (natural-log lnnormal noise), which this problem also uses: #509 is fixed and now lets Blasi import; this steady-state gap is what stops it at config load.

PyBNF version: 1.6.0 (main).