Port convolutions to cuDNN v8 API

[mk-61]

Description

This change ports Convolution and Deconvolution operations to cuDNN v8 API.
Legacy API support is dropped, as per this RFC: #20618.

The change also includes some cuDNN v8 API general support stuff, to be re-used later when more operations are ported to the v8 API.

Finally, auto-tuning functionality is moved from cuDNN into MXNet, hence some memory management changes were required.

Checklist

Essentials

• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented

[mxnet-bot]

Hey @mk-61 , 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: [website, sanity, unix-cpu, windows-cpu, edge, windows-gpu, centos-cpu, centos-gpu, unix-gpu, clang, 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.

[mk-61]

@mxnet-bot run ci [miscellaneous, unix-gpu]

[mxnet-bot]

Jenkins CI successfully triggered : [unix-gpu, miscellaneous]

[mk-61]

@mxnet-bot run ci [unix-gpu]

[mxnet-bot]

Jenkins CI successfully triggered : [unix-gpu]

[mk-61]

@mxnet-bot run ci [unix-gpu]

[mxnet-bot]

Jenkins CI successfully triggered : [unix-gpu]

[DickJC123]

Thanks again for your responsiveness to our review comments, and patience throughout these recent CI issues (e.g. #20738). The problem with the apt repo server seems to have been solved, and this PR has passed without new work-arounds added. LGTM.