Update dependency mkl_dnn to v0.21.5

[optible-bot]

This PR contains the following updates:

Package	Type	Update	Change
mkl_dnn	http_archive	patch	v0.21.3 -> v0.21.5

---

Release Notes

oneapi-src/oneDNN

v0.21.5

Compare Source

This is a patch release containing following changes to v0.21.4:

• Fixed s8 reorders that did not compute compensation correctly (d446661, 7a49772)
• Fixed potential buffer overflow in int8 convolution scratchpad (8c5c7cf)
• Fixed segfault for s8 reorders on blocked formats (9497acc, 6f1d0c9)
• Fixed correctness in fp32 convolution weight gradient with dilation and padding (503bf57, d00afab)
• Fixed correctness inssue in 1D bfloat16 dilated convolution (481dd39)

v0.21.4

Compare Source

This is a patch release containing following changes to v0.21.3:

• Fixed large padding handling in input tensor transposition in bfloat16 weights gradient convolution (6df67fe)
• Fixed performance of reference convolution (2e1d048)
• Fixed "code is too big" error in case of extreme large spatial size (ed0be61, 4dee389, 59759ba)

---

Configuration

📅 Schedule: At any time (no schedule defined).

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, click this checkbox.

---

This PR has been generated by Renovate Bot.