feat: add multi-engine regex flavour support

docs/ARCHITECTURE.md

@@ -9,13 +9,22 @@ React workbench
   │    ├─ Regexpp ECMAScript provider → normalized AST
   │    ├─ PCRE2 lexical provider → partial normalized structure
   │    ├─ Python lexical provider → partial normalized structure
-  │    └─ Java lexical provider → partial normalized structure
+  │    ├─ Java lexical provider → partial normalized structure
+  │    └─ PHP/Perl/Ruby/C++/Go/Rust/.NET/Scala lexical providers → partial structures
   ├─ Replacement SyntaxSupervisor → syntax.worker → typed replacement tokens
   ├─ EngineSupervisor
   │    ├─ ecmascript.worker → native RegExp → match DTOs
   │    ├─ pcre2.worker → PCRE2 10.47 WASM ABI → match DTOs
+  │    ├─ php.worker → PHP 8.5.8 preg / PCRE2 10.44 → match DTOs
+  │    ├─ perl.worker → legacy WebPerl 0.09-beta / Perl 5.28.1 → match DTOs
   │    ├─ python.worker → Pyodide 314.0.3 → CPython 3.14.2 re → match DTOs
-  │    └─ java.worker → TeaVM 0.15.0 java.util.regex module → match DTOs
+  │    ├─ ruby.worker → ruby.wasm → CRuby 4.0.0 Regexp → match DTOs
+  │    ├─ java.worker → TeaVM 0.15.0 java.util.regex module → match DTOs
+  │    ├─ cpp.worker → Emscripten 6.0.4 libc++ std::wregex → match DTOs
+  │    ├─ go.worker → Go 1.26.5 standard-library regexp → match DTOs
+  │    ├─ rust.worker → Rust regex crate 1.13.1 → match DTOs
+  │    ├─ dotnet.worker → .NET 10.0.10 System.Text.RegularExpressions → match DTOs
+  │    └─ scala.worker → explicit compatibility adapter → TeaVM Java module
   ├─ Pcre2TraceSupervisor → pcre2-trace.worker
   │    └─ separate ABI-v3 PCRE2_AUTO_CALLOUT operation → bounded trace DTOs
   ├─ AnalysisPanel
@@ -50,6 +59,9 @@ a new generation. Timeout is a distinct error state. A separate idempotent
 load operation verifies a worker's runtime identity before execution; its
 flavour-specific startup deadline is not charged to the 250 ms live or selected
 manual execution deadline. A killed generation must load and verify again.
+The supervisor has a two-worker retention target; before creating another
+profile worker it evicts the least recently used idle generation. Active work
+is never evicted and may temporarily exceed that target.
 
 The Python worker imports only the self-hosted Pyodide loader and pack, then
 evaluates the application-owned bounded bridge in CPython. The bridge returns
@@ -59,6 +71,22 @@ ES2015 module. That module executes TeaVM's own `java.util.regex.Pattern` and
 `Matcher` class-library implementation and returns native UTF-16 ranges; it is
 not OpenJDK.
 
+PHP, Perl, Ruby, C++, Go, Rust and .NET each use a fixed bounded bridge to the
+named self-hosted runtime. User values cross as structured/JSON data, never as
+generated source. Scala intentionally reuses the TeaVM Java module and rewrites
+only request/result profile metadata; it contains no Scala runtime. The legacy
+Perl topology is unusual: the supervised module worker owns the classic worker
+required by WebPerl, so supervisor termination still discards the complete
+worker tree.
+
+Ruby's production bridge avoids ruby.wasm's generic JavaScript object
+conversion: the worker writes one bounded length-prefixed request into its
+private WASI memory filesystem, calls the fixed bridge with no user arguments,
+reads one bounded inert metadata JSON string, and obtains bounded UTF-8
+replacement output from a separate private file. That path avoids JSON output
+amplification and passes the same strict-CSP browser gate as the other
+profiles.
+
 PCRE2 tracing is not an `EngineSupervisor` match or replacement operation. Its
 own worker compiles with `PCRE2_AUTO_CALLOUT`, copies only complete fixed
 records and bounded mark bytes, and stops natively at either trace cap.
@@ -86,9 +114,10 @@ deterministic golden identities and an exact PCRE2 10.47 compile/execute gate.
 
 Subject minimization is a main-thread bounded reducer whose expensive predicate
 checks stay in independently terminable ECMAScript or PCRE2 workers. Fixed
-pattern syntax is parsed once for capture metadata. Python and Java are blocked
-at this feature boundary rather than routed to a different engine. The baseline
-and every accepted candidate retain the exact target and engine identity.
+pattern syntax is parsed once for capture metadata. Every other profile is
+blocked at this feature boundary rather than routed to a different engine. The
+baseline and every accepted candidate retain the exact target and engine
+identity.
 Timeout, crash, cancellation, worker failure and incomplete output are separate
 outcomes. Deterministic ddmin deletion is followed by a fixed-point local sweep
 over Unicode-scalar deletion and lower-rank canonical replacement; only
@@ -139,13 +168,13 @@ separator during apply. This makes anchor and cross-line behavior deliberately
 different rather than silently rewriting the pattern. Match DTOs are reduced to
 bounded per-group participation counts and samples before the next document.
 
-Native ranges are retained in the engine's declared unit. ECMAScript and Java
-use UTF-16 code units. PCRE2 always uses UTF/UCP and exposes UTF-8 byte ranges;
-its ABI copies scalar records and names out of WebAssembly before returning and
-retains no code pointer. CPython `re` exposes Unicode code-point indices.
-Browser editor ranges are half-open UTF-16 code-unit ranges. Reusable
-converters cover UTF-8 byte and Unicode code-point offsets, including
-invalid-boundary rejection and lone-surrogate detection.
+Native ranges are retained in the engine's declared unit. ECMAScript, Java,
+Scala compatibility and .NET use UTF-16 code units. Standalone PCRE2, PHP, Go
+and Rust expose UTF-8 byte ranges. CPython, CRuby, libc++ `std::wregex` and
+legacy Perl expose Unicode code-point positions. Browser editor ranges are
+half-open UTF-16 code-unit ranges. Reusable converters cover UTF-8 byte and
+Unicode code-point offsets, including invalid-boundary rejection and
+lone-surrogate detection.
 
 The release boundary is `dist/` plus checked-in legal/source documents. The
 Portal consumes the resulting immutable ZIP and never imports React source.