Tracking: migrate from sciline map/reduce to stages and aggregations (sciline ADR 0003)

[SimonHeybrock]

Tracking issue for the ESS side of sciline ADR 0003, Replace map/reduce with stages composed outside the graph, proposed in scipp/sciline#245. The rollout plan in the design document has the file-level sites for every item below, from a survey of the monorepo and esslivedata on 2026-09-11. Links point at the sciline branch until the ADR PR merges. Nothing here starts before the ADR is accepted and the additive sciline release is out.

In short: a Stage is the part of a pipeline from input keys to output keys with the static part held; an Aggregation is two stages with an accumulator per accumulation key between them. Each package returns its own object in place of a map/reduced pipeline, holding the pipeline, its aggregations, and the contributions by member.

ess.reduce, additive

• Accumulators: maybe_hist moves out of Accumulator.push into the subclasses that histogram; a Forwarder; a test that every accumulator satisfies sciline.Accumulator.
• StreamProcessor rewritten as a driver over three stages, a forwarder, and accumulators, keeping its interface, against the streaming, visualize, and accumulator tests. Needs the sciline follow-up (provide, reporter, visualize support).
• assign_parameter_values and parameter_mappers replaced by the package-object protocol, with or after esssans.
• docs/user-guide/polarization/zoom.ipynb (uses get_mapped_node_names and with_sample_runs) migrates with esssans.

Reduction packages, one PR each, in this order

• esssans: with_sample_runs/with_background_runs become one aggregation per run type with a shared finalize stage held by a package object; with_pixel_mask_filenames becomes a PixelMaskFilenames list parameter and two providers; with_banks becomes compute_members; ZoomTransmissionFractionWorkflow; parameter_mappers registrations; LoKI and zoom notebooks; tests using compute_mapped. Sets the package-object pattern.
• essdiffraction: with_pixel_mask_filenames becomes a list parameter and one provider and loses the empty-list cyclebane workaround. All 22 test call sites pass [], so add one test with a mask file. The two DREAM advanced-notebook tables (one with two columns) become an aggregation.
• essspectroscopy: three bank folds in the bifrost builders. Open design point for that PR: NeXusData depends on the run, so the bank aggregation sits inside per-run work when runs are members; either the package object owns a per-run bank aggregation or a DetectorNames list parameter with a looping provider. compute_mapped in one test and one notebook becomes compute_members.
• essreflectometry: with_filenames becomes an aggregation with up to seven accumulation keys and loses the try/except around each reduce; BatchProcessor.compute loses its compute_mapped fallback; batch_processor enters the aggregation for a list-valued Filename[SampleRun]; the GUI folds; amor and estia notebooks; tests.
• essnmx: notebooks and one test fixture only; mcstas_workflow.ipynb imports cyclebane.graph directly and must stop.
• essimaging: drop visualize(compact=) in two notebooks.

Every package also drops visualize(compact=) from its notebooks in its own PR. constraints= (essreduce nexus workflows, offspec, amor) stays until the breaking sciline release.

esslivedata

• The bifrost bank fold in config/instruments/bifrost/factories.py sits in the static part of the streaming pipeline; it becomes Aggregation(...).compute(...) with the result set as a parameter before the StreamProcessor is built. Consumes the rewritten StreamProcessor through the essreduce bump.
• Drop the direct cyclebane>=26.9.0 pin once sciline drops cyclebane.

Then

• essreduce raises its sciline minimum to the breaking release; offspec and amor drop constraints= in the same bump.
• essapps: apply the edits listed at the end of stages.md on branch architecture-sketch, reading Fold as Aggregation and "accumulation point" as accumulation key.