[BUGFIX] Optimize preparation of self attention operators

[bgawrych]

Description

This PR fixes problem with redundant double reorder which occurs when dealing with DNNLData

[mxnet-bot]

Hey @bgawrych , Thanks for submitting the PR
All tests are already queued to run once. If tests fail, you can trigger one or more tests again with the following commands:

• To trigger all jobs: @mxnet-bot run ci [all]
• To trigger specific jobs: @mxnet-bot run ci [job1, job2]

---

CI supported jobs: [windows-cpu, clang, website, sanity, centos-cpu, unix-gpu, centos-gpu, unix-cpu, edge, windows-gpu, miscellaneous]

---

Note:
Only following 3 categories can trigger CI :PR Author, MXNet Committer, Jenkins Admin.
All CI tests must pass before the PR can be merged.

[szha]

Is there a test that covers the modified code path?

[bgawrych]

Is there a test that covers the modified code path?

@szha There is a test for these operators https://github.com/apache/incubator-mxnet/blob/75e4d1d3e41cc5aef9a0141e40127a75e02ccd04/tests/python/dnnl/subgraphs/test_matmul_subgraph.py#L33

These change only removes redundant data reorder