feat: add multi-engine regex flavour support

docs/EXTRACTION_MODEL.md

@@ -18,14 +18,22 @@ budget. Consumers must inspect the preview status rather than treating a
 preview as complete. In particular, list export is disabled if its template
 would consume an incomplete value.
 
-Repeated ECMAScript, PCRE2, Python and Java captures expose only the final
-retained capture. The UI labels this limitation and does not invent history.
-Later .NET support may add actual engine capture-history nodes without
-degrading them to this model.
+Repeated captures expose only the final retained value in all twelve v0.4.0
+profiles. This includes .NET even though its native API can retain capture
+history: the current adapter deliberately normalizes only the final record and
+reports `captureHistory: false`. The UI labels the limitation and does not
+invent or silently discard advertised history.
 
-The syntactic hierarchy remains useful even if actual spans overlap or fall
-outside a parent's geometry; native and normalized ranges are preserved rather
-than “fixed”. ECMAScript and Java native offsets are UTF-16 code units, PCRE2
-native offsets are UTF-8 bytes and Python native offsets are Unicode code
-points. All editor ranges are normalized to UTF-16 without discarding the
-native unit.
+The partial syntactic hierarchy remains useful when the active provider found
+capture relationships, even if actual spans overlap or fall outside a parent's
+geometry. For C++ alternate grammars or suppressed submatches, capture
+discovery can be explicitly unavailable. Runtime records remain authoritative.
+
+Native and normalized ranges are preserved rather than “fixed”:
+
+- JavaScript, Java, Scala compatibility and .NET use UTF-16 code units;
+- standalone PCRE2, PHP, Go and Rust use UTF-8 bytes; and
+- Python, Ruby, C++ and legacy Perl use Unicode code points.
+
+All editor ranges are normalized to UTF-16 only after the adapter verifies
+valid boundaries. The native unit and original range remain available.