TypeSchema: store slicing separately from fields

[ryukzak]

First of a two-PR sequence restructuring how TypeSchema stores and processes slice information.

• Store slicing separately from fields: slicing: Record<fieldName, FieldSlicing> at schema level (specializations, profiles, snapshots, nested types); fields describe the data shape, slicing is an independent constraint layer. Keyed by element name, so choice declarations/instances and regular fields are covered uniformly.
• flatProfile merges slicing maps leaf-wins per field (mirroring the field merge); tree shaking prunes slicing entries for removed fields.
• Own slicing vocabulary: SliceDiscriminator + inline rules union replace FS.FHIRSchemaDiscriminator/FS.SlicingRules; the repeated type-discriminator checks collapse into the isTypeDiscriminated predicate.

Generated TS/Python output is byte-identical; only the serialized TypeSchema shape changes:

// before
"fields": {
  "category": { "type": ..., "array": true, "slicing": { "discriminator": [...], "slices": {...} } }
}

// after
"fields":  { "category": { "type": ..., "array": true } },
"slicing": { "category": { "discriminator": [...], "slices": {...} } }

Follow-up PR (branch typeschema-slice-derived-facts, stacked on this one): precomputed derived slice facts on FieldSlice (effectiveRequired, constrainedChoice, autoStub, resourceType), deduplicating the TS/Python writer collectors and fixing the divergences that duplication produced.