Implement support for `TritonGPU::UpcastMXFPOp` for Intel XPU BE

[etiotto]

Triton introduced a new operation named UpcastMXFPOp in the TritonGPU dialect.
This operation was designed to compute the bf16 encoded in a given mxfp number as per: https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf

[etiotto]

As discussed offline @LiyangLingIntel will work on the lowering code for upcast_msfp while I work on the feature to decompose the tt.dot_scaled operator into "upcast_mxfp + dot".

[leonling-ll]

Created #2961 to track the following task that support dot operand encoding for UpcastMXFPOp.