toFHIR creates duplicate Patient resources when Person data is split across multiple COMPOSITION/content[n] entries (multiple=true)

[SevKohler]

Body

Summary

When mapping KDS_Person from openEHR to FHIR with multiple=true, the engine may produce multiple Patient resources for a single logical patient if mapped fields come from different COMPOSITION/content[n] roots.

Current Behavior

OpenEhrToFhir groups created resources by a key derived from the first openEHR index of each mapped path.
If mappings resolve from different content[n] entries (e.g. ADMIN_ENTRY.person_data.v0, EVALUATION.gender.v1, ADMIN_ENTRY.versicherungsinformationen.v0), separate Patient instances are created.

Expected Behavior

All Person-related mappings for one composition should merge into one Patient (unless the model explicitly represents multiple patients).

Impact

• Partial duplicate Patient resources in output bundle.
• Users currently need multiple: false as workaround in Person mapping models.

Reproduction

1. Use Person context/model with:

• ADMIN_ENTRY.person_data.v0 as start
• extensions mapping additional patient fields from other content[...] archetypes (gender, insurance).

2. Run toFHIR on composition containing those archetypes in separate content[n].
3. Observe bundle contains 2+ Patient entries (one partial, one fuller).

Root Cause (technical)

Resource creation key is index-based (derived from openEHR path index) when fhirConfig.multiple=true, so logically-related data from different content[n] does not merge.

Proposed Fix

Introduce a stable logical grouping strategy for toFHIR resource assembly, instead of using only openEHR index:

• group by primary target resource/root mapping context, and/or
• allow explicit grouping key in mapping config.

Keep index-based behavior only where truly needed for repeated resource instances.

Workaround

Set spec.fhirConfig.multiple: false in Person model (ADMIN_ENTRY.person_data.v0) to force single Patient.

Acceptance Criteria

• Given Person data split across multiple content[n], toFHIR returns exactly one Patient.
• Existing genuinely multi-instance models remain supported.
• Add regression test covering split-content Person mappings.

NOTE:

This might be fixed with the introduction of the composition layer as it is planned.