fix(validator): type Element.id-derived ids as string, not id (#424)

[mauripunzueta]

Closes #424.

Problem

The R4B and R5 schema packs typed ElementDefinition.id (and every
Element.id-derived id) as the FHIR primitive id, whose regex
[A-Za-z0-9\-\.]{1,64} rejects element ids that contain [x] (choice
elements), : (named slices), or exceed 64 characters. Any
StructureDefinition carrying a choice element or a named slice therefore
failed validation — including StructureDefinitions the FHIR spec itself
publishes (e.g. ActivityDefinition.subject[x]). R4 and R6 were already
correct.

Root cause

In the converter, not the packs. EdType::effective_code() resolves a
FHIRPath System.* code through the structuredefinition-fhir-type
extension's valueUrl. HL7 populates that hint inconsistently:

element	base	R4 ext	R4B ext	R5 ext	R6 ext	fhir.schema.json
ElementDefinition.id	Element.id	string	id	id	string	string (all)
Element.id / Address.id	Element.id	string	id	mixed	string	string (all)
Patient.id / Account.id	Resource.id	string	id	id	string	id (R4/R5/R6)
Extension.url	Extension.url	uri	uri	uri	uri	uri (all)

The authoritative type is the base element: FHIR defines Element.id
as string and Resource.id as the constrained id token, and HL7's own
fhir.schema.json types every element id as string, every resource id as
id, and Extension.url as uri, in all four versions.

Fix

The converter now keys the value-domain type on base.path: an element
deriving from Element.id resolves to string, overriding a spurious id
extension. Resource.id and Extension.url continue to honor the
extension.

This was chosen over the issue's option 1 ("map every System.String →
string"), which would have relaxed resource-id validation
(Patient.id allowing spaces / >64 chars) and demoted Extension.url to
string — both contradicting fhir.schema.json. The base-driven rule
fixes the whole class without those regressions.

Effect on the regenerated packs

• R4B / R5: every Element.id-derived id flips id→string (60 and
  73 elements). Resource ids stay id; Extension.url stays uri.
• R4 / R6: element ids were already string; the only change is 5
  ViewDefinition backbone .id fields (select.id, where.id,
  constant.id, select.column.id, select.column.tag.id), which are
  Element.id-derived and carried the same spurious id extension. These
  are genuine instances of the same bug; ViewDefinition.id (the resource
  id) is untouched.

Scope / DB

Entirely within helios-fhir-validator (converter + embedded packs).
Validation is backend-agnostic and the packs are shared across all storage
backends, so there are no per-backend / no SQL code paths to change.

Tests

• Converter unit tests pin the rule: Element.id→string,
  Resource.id→id, Extension.url→uri, and that the override is
  independent of the (inconsistent) extension value.
• An R4B/R5 end-to-end test (pack_smoke, #[ignore]d like its siblings)
  validates a StructureDefinition whose element ids carry [x] and : and
  asserts no primitive-value error on any .id path. Pre-fix this
  produced two such errors per version; post-fix, zero.
• Full helios-fhir-validator suite green; fmt and clippy clean.

Notes

• Out of scope: Validator: FHIR string/markdown regex rejects non-breaking spaces on R4 and R4B (Unicode-aware \S) #425 (nbsp vs Unicode \S in the string regex) — the
  string regex is deliberately untouched.
• Verifiable against test(validator): run the pack tests in CI, sweep the official FHIR example corpus #423's corpus sweep: the R4B/R5 primitive-value
  cluster attributable to this bug drops out; that PR's baselines will need
  regenerating once both land.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!