Improve README: FHIR R4-R6, simplify bulk match, clarify client-driven merge

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: skylightis666

docs/README.md

@@ -10,13 +10,13 @@ MDMbox is a master data management service for healthcare organizations. It iden
 
 **Probabilistic matching.** Configurable Fellegi-Sunter models compare records across multiple dimensions (name, date of birth, address, phone) and produce a match score. Handles typos, incomplete data, and transpositions.
 
-**Merging.** Client-driven merge with a FHIR transaction Bundle. Full audit trail via Task and Provenance resources. Preview mode for dry runs.
+**Merging.** FHIR R5-aligned `$merge` operation with a client-driven approach -- the client builds the merge plan as a FHIR transaction Bundle, giving full control over which fields survive and how related resources are reassigned. Full audit trail and preview mode.
 
-**Bulk matching.** Pre-materialize data into a flat table, then run parallel workers to find all duplicate pairs across millions of records. Download results as CSV.
+**Bulk matching.** Find all duplicate pairs across millions of records in parallel. Download results as CSV.
 
 **Admin UI.** Server-rendered interface for managing matching models and running bulk match jobs.
 
-**FHIR R4.** All operations use standard FHIR resource types and follow FHIR conventions. Works with any resource type (Patient, Practitioner, Organization, etc.).
+**FHIR R4 to R6.** Matching, merging, and referencing operations work with any FHIR resource type -- Patient, Practitioner, Organization, or any other. Configure a matching model for the resource type you need.
 
 ## Deployment modes