[tanvir-ux]

Summary

PipedInputStream and PipedOutputStream are Closeable, but their constructors were not modeled, so Pulse never tracked them as resources. A pipe end that is only closed on the normal path was not reported when a read or write throws. The existing tests FN_pipedInputStreamNotClosedAfterReadBad and FN_pipedOutputStreamNotClosedAfterWriteBad in ResourceLeaks.java show this.

• Model the <init> of both classes with Resource.allocate ~exn_class_name:"java.io.IOException", since the constructors that connect to the other end of the pipe may throw IOException (same pattern as Socket/ServerSocket).
• read, write, flush and close already go through the existing InputStream, OutputStream and Closeable models, so nothing else is needed.

Tests

• FN_pipedInputStreamNotClosedAfterReadBad and FN_pipedOutputStreamNotClosedAfterWriteBad are renamed without the FN_ prefix and are now reported.
• pipedInputStreamClosedOk and pipedOutputStreamClosedOk closed the stream in finally without a null check. With the constructor able to throw, Pulse correctly flags that as a null dereference, so I added if (... != null) before close(), as the FileInputStream tests already do. They stay clean.

Test plan

• Before (main @ 333eda4): neither piped-stream test is reported
• After: PULSE_RESOURCE_LEAK for both, allocated by the PipedInputStream() / PipedOutputStream() constructors; the Ok cases stay clean
• make direct_java_pulse_test: the only issues.exp changes are those 2 lines. (Locally I still get the unrelated lambda-name differences in InvokeDynamic/TransitiveAccess that also show up on main with my javac; not committed.)
• ocamlformat 0.29.0 reports no changes for PulseModelsJava.ml