chore(release): v0.9.0 - SGLang-driven serving hardening + OOM crash guard

Highlights since v0.8.1:
- response_format: clean schema-shaped JSON on both streaming and
  non-streaming, Gemma4 AND Qwen3.6 (first-JSON-value truncation +
  Eager schema grammar). Fixes silent free-text on non-streaming.
- Tool-call grammar enforcement: tool_choice=required/named now
  structurally fills required params on quantized Gemma4 + Qwen3.6
  (kills the empty-parameter infinite-loop defect). Plus a Qwen
  vocab-padding fix that makes its grammar actually enforce.
- Robustness: prompt-size reject cap now guards ALL MLX chat paths
  (OpenAI + Anthropic, stream + non-stream) so an oversized prompt
  returns a clean error instead of crashing the whole server via an
  uncaught Metal OOM. Safer default (16384).
- GET /v1/loads JSON observability; multi-entry LPM prefix cache for
  branching agentic sessions; MLX native decode overlap.
- App: memory calculator with model-aware tuning + budget auto-optimize.

Verified on Gemma-4-26b-nvfp4 + Qwen3.6-35B-mxfp4 (e2e suite 10/10, 8/8).

Author: rabbitson87

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "lumen-app"
-version = "0.8.1"
+version = "0.9.0"
 edition.workspace = true
 rust-version.workspace = true
 license.workspace = true

crates/lumen-app/Cargo.toml

@@ -1,7 +1,7 @@
 {
   "name": "lumen-app-frontend",
   "private": true,
-  "version": "0.8.1",
+  "version": "0.9.0",
   "type": "module",
   "scripts": {
     "predev": "cargo build --manifest-path ../../../Cargo.toml -p lumen-server --release",

crates/lumen-app/frontend/package.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "Lumen",
-  "version": "0.8.1",
+  "version": "0.9.0",
   "identifier": "ai.lumen.app",
   "build": {
     "frontendDist": "frontend/dist",