[SPARK-56447][SHELL] Fix spark-shell REPL classpath initialization error

[jiwen624]

What changes were proposed in this pull request?

In Main.doMain, prepend java.class.path to the -classpath argument passed to the REPL's GenericRunnerSettings:

Why are the changes needed?

$SPARK_HOME/bin/spark-shell fails with:

Failed to initialize compiler: object scala in compiler mirror not found.
** Note that as of 2.8 scala does not assume use of the java classpath.
** For the old behavior pass -usejavacp to scala, or if using a Settings
** object programmatically, settings.usejavacp.value = true.
Exception in thread "main" java.lang.NullPointerException: Cannot throw exception because "null" is null
        at scala.tools.nsc.CompilationUnits$CompilationUnit.<init>(CompilationUnits.scala:43)
        at scala.tools.nsc.CompilationUnits$CompilationUnit.<init>(CompilationUnits.scala:44)
        at scala.tools.nsc.interpreter.IMain$ReadEvalPrint.compile(IMain.scala:734) 

This works in previous versions.
bin/spark-shell has passed -Dscala.usejavacp=true since SPARK-4161 to address exactly this behaviour. The mechanism is broken after SPARK-52587 which initializes the compiler before the classpath is set. The fix makes doMain explicitly include java.class.path in the -classpath argument, restoring correct behaviour regardless of how SparkILoop is constructed.

Does this PR introduce any user-facing change?

Yes. spark-shell was broken with a compiler initialization error. This PR fixes it so spark-shell starts and evaluates
expressions correctly again.

How was this patch tested?

Added a unit test in ReplSuite that calls Main.doMain with Array.empty (no explicit -classpath argument) and verifies the output. This test fails on the unfixed code and passes after the fix.

Was this patch authored or co-authored using generative AI tooling?

No