[GPU] Fixup GEMM tail handling

[kealan-barbieri]

Description

Fixup handling tails for weights decompression GEMM cases:

• reenable NaN detection in fp8->bf emulated conversion routine, was hiding issues
• disable kDescRem remainder handling parameter in cases where it causes overreading garbage data that may not be masked out by multiplying by 0.

Fixes # MFDNN-14869

Checklist

General

• Do all unit and benchdnn tests (make test and make test_benchdnn_*) pass locally for each commit?
• Have you formatted the code using clang-format?

Bug fixes

• Have you included information on how to reproduce the issue (either in a github issue or in this PR)?
• Have you added relevant regression tests?

[kealan-barbieri]

make test
set test_scope=NIGHTLY
disable test_device_cpu
enable test_device_gpu
disable benchdnn_all
enable benchdnn_matmul
enable benchdnn_ip
enable benchdnn_rnn
enable benchdnn_graph
enable benchdnn_lrn