BEGIN_PUBLIC
[Kernel][GPU] Add MAX compatibility lane support

Implement the MAX tensor compatibility work needed by modular#6076
and modular#5300, including composed tensor helpers, repeat and reverse semantics,
nearest resize wiring, bitwise ops, mixed integer-bfloat16 multiply parity,
and the related graph, eager, and interpreter tests.

GPU reruns for the updated eager repeat_interleave path are still pending;
RunPod exited pods during Bazel external fetch and setup after the code
changes, so the remaining gap is validation infrastructure rather than a
localized product failure.
END_PUBLIC

Signed-off-by: Omar Habra <omarbro4all@gmail.com>