[examples] Remove obsolete kernel dump levels from XeGPU matmul example

[tkarna]

We are now using the monolithic gpu-lower-to-xevm-pipeline pass to lower below the xegpu workgroup level, so the lowering cannot be stopped on subgroup/instruction/etc levels anymore.

While having such diagnostics is useful for debugging, to support it we'd need to replicate the xevm lowering pipeline in lighthouse as a schedule implying some sync/maintenance burden.