Fix extraction progress caching and JVM tuning

- Replace Map-based archive item cache with plain Object.create(null)
  to work around mysterious Map.has() returning false despite set()
  being called with the same key — this caused resolveArchiveItems
  to run on every 1.1s pulse instead of being cached, preventing
  extraction progress (Entpacken X%) from ever showing in the UI
- Apply same fix to both hybrid and full extraction paths
- Increase JVM heap from 512MB to 1GB for better extraction throughput
- Use SerialGC for faster JVM startup on short-lived extract processes
- Add download lifecycle logging (package add + item download start)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/main/extractor.ts

@@ -1019,8 +1019,9 @@ function runJvmExtractCommand(
   const args = [
     "-Dfile.encoding=UTF-8",
     `-Djava.io.tmpdir=${jvmTmpDir}`,
-    "-Xms32m",
-    "-Xmx512m",
+    "-Xms64m",
+    "-Xmx1g",
+    "-XX:+UseSerialGC",
     "-cp",
     layout.classPath,
     JVM_EXTRACTOR_MAIN_CLASS,