xe: increase concat lws[0] size w/new optimal calculation

[syurkevi]

Description

This PR adresses performance regressions discovered as part of regular monitoring by increasing the lws[0] as much as possible. The new get_optimal_lws() logic is biased towards a more n-dimensional lws. Removing the requested vectorization axis (-1) will create a lws[0] as large as possible. If maximizing lws[0] will cause new slowdowns an updated reverted get_optimal_lws() will be used instead.

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?

Performance improvements

• Have you submitted performance data that demonstrates performance improvements?
  

[syurkevi]

make test perf-gpu
set primitive=concat