Make URLJarCollector tolerant of JDK 21 (and newer JVMs)

[pavelsavara]

Summary

This change hardens the static initializer in
robocode.core/src/main/java/net/sf/robocode/io/URLJarCollector.java so Robocode
runs correctly on JDK 21 and other modern JVMs where the internal
sun.net.www.protocol.jar.* layout differs from the classic Sun/Oracle JVM.

Problem

URLJarCollector reflects into the JDK-internal classes
sun.net.www.protocol.jar.JarFileFactory and
sun.net.www.protocol.jar.JarURLConnection to enable a fast, caching JAR
connection path. On JDK 21:

• The reflective access into these internal classes can fail with a
  RuntimeException (e.g. InaccessibleObjectException from the module system),
  which was not caught by the previous catch clause.
• Even when the fields resolve, fileCache / urlCache may not be in the shape
  the caching path assumes.

Because the previous code only caught the checked reflection exceptions
(ClassNotFoundException | NoSuchFieldException | IllegalAccessException), any
RuntimeException thrown during initialization escaped the static block and
prevented the class from loading.

Fix

• Treat the JVM as a "classic Sun/Oracle JVM" (sunJVM = true) only when both
  fileCache and urlCache were successfully resolved to non-null values,
  rather than assuming success after the reflection calls.
• Also catch RuntimeException so unexpected failures (such as the module-system
  access errors seen on JDK 21) gracefully fall back to the non-caching code path
  instead of failing class initialization.

When the JVM is not recognized, openConnection simply disables URL caching
(urlConnection.setUseCaches(false)), which is the correct, portable behavior.