[ESSSPECTROSCOPY] Source the BIFROST Bragg peak monitor from the NXmonitor

[SimonHeybrock]

The BIFROST Bragg peak monitor workflow could not run against any real file. get_calibrated_bragg_peak_detector required NeXusComponent[NXdetector], but the Bragg peak monitor is the elastic monitor (cbm5), written as an NXmonitor in CODA files, in the geometry artifacts and in the McStas simulation file alike, so loading raised

ValueError: The NeXus group 'elastic_monitor' was expected to be a NXdetector but is a NXmonitor

Only the user guide worked, and only because it stands a detector triplet in for the monitor — it carries an explicit warning saying so.

The monitor's component, transformation and event data now come from ElasticMonitor. The workflow already declares monitor_types=(ElasticMonitor, NormalizationMonitor), so every loading node exists and this is a re-pointing rather than new machinery. A monitor has no pixel offsets, so the position is the transformed origin as in get_calibrated_monitor, and no detector_number, so the single pixel is named explicitly for event assembly. The Analyzer dependency is dropped: get_base_calibrated_detector_bifrost never reads it, and a monitor in the direct beam has no analyzer.

Event assembly no longer goes through assemble_detector_data, which groups by event_id. Streamed monitor events carry none — the file-based path hid this because CODA's cbm5_events does have event_id. It was only found by running the live path.

Along the way this fixes a latent defect that is not specific to the monitor: a time-dependent position makes ltotal carry time, but group_by_rotation has already renamed that dimension to a4, so the broadcast in detector_wavelength_data rejected it. This bites whenever the tank rotates. It stayed hidden because the user guide's stand-in position is static, and because the inelastic/Q-cut path uses a different detector_wavelength_data that takes no ltotal at all.

Testing

18/18 essspectroscopy tests pass and the user guide still computes (its EmptyDetector stand-in moved to simulation_providers).

There is no regression test for the monitor path here: the simulated data contains no Bragg peak monitor with event data — the simulation file's elastic_monitor is histogram-mode. The path is instead exercised end to end downstream in scipp/esslivedata#1236, where streamed monitor events plus the live rotation readbacks produce a populated (Qpar, Qperp) map. That branch does not work against released essspectroscopy, so it is blocked on this PR.

Open questions for reviewers

The scientists asked for the map in the laboratory frame, but project_momentum_transfer projects sample_table_momentum_transfer — the sample-table frame, which is what you would want to compare against an expected reciprocal lattice. Worth settling before merge.

single_crystal/detector.py imports _assign_detector_position from the sibling ..detector; intra-package but private. Say the word if you want it promoted.