Fix: Wrong handle in multi-stream handle test

[sjain-stanford]

The multi-stream handle test failed flakily on a recent CI run on main: https://github.com/iree-org/fusilli/actions/runs/22334529677/job/64623964052

-------------------------------------------------------------------------------
Handle creation with stream and deviceId
  Create multiple handles with different HIP streams
-------------------------------------------------------------------------------
/workspace/tests/hip_tests/test_handle.cpp:92
...............................................................................

/workspace/tests/hip_tests/test_handle.cpp:136: FAILED:
  REQUIRE( readData == hostData )
with expansion:
  { 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.
  0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
  0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f }
  ==
  { 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.
  14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.
  14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f, 3.14f }

terminate called after throwing an instance of 'std::bad_variant_access'
  what():  std::get: wrong index for variant
/workspace/tests/hip_tests/test_handle.cpp:136: FAILED:
  {Unknown expression after the reported line}
due to a fatal error condition:
  SIGABRT - Abort (abnormal termination) signal

• Fixed copy-paste bug in test_handle.cpp where the "Create multiple handles with different HIP streams" test was using handle1 instead of handle2 for the second buffer's allocate and read calls
• (I'm not sure if this is accurate) This could have caused a cross-stream race condition resulting in flaky D2H reads returning all zeros instead of the expected 3.14f values

🤖 Generated with Claude Code

[AaronStGeorge]

This could have caused a cross-stream race condition resulting in flaky D2H reads returning all zeros instead of the expected 3.14f values

The fix is definitely what we want, but I'm a little lost as to why it would cause a race condition. Everything should have been stream ordered on handle 1.

[sjain-stanford]

The fix is definitely what we want, but I'm a little lost as to why it would cause a race condition. Everything should have been stream ordered on handle 1.

Hence my prefix on what Claude generated :P
(I'm not sure if this is accurate)