[AWS] Fix interrupt mappings which are set to a bad default after certain ENI operations

[Philip Cox]

SRU Justification:

[Impact]
Interrupt mapping is set to a bad default after certain ENI operations

After using ethtool to update the (rx) ring buffer settings on an ENI, the interrupt CPU affinity for all tx-rx queues is reset to a bad default - CPU 0-31 - even if the ENI is on a different CPU socket (NUMA node). Another operation that causes the interrupt mapping to be reset to CPU 0-31 is moving an interface to a network namespace.

[Fix]

ENA driver version 2.12.3g. They mentioned that this commit in the ENA code looks like it addresses the bug: https://github.com/amzn/amzn-drivers/commit/ed7754cdf6ca0d4fb3d76906abc17718cc3ffa23

It's included starting in driver version 2.12.0.

I have back ported the patch from the AWS OOT driver, as well as any existing prerequisites needed.

[Test Plan]

I have tested this, as has AWS.

[Where problems could occur]

As this is an Ubuntu Sauce patch, if these changes make their was into the stable trees, or any patches touching this area of code, they may cause merge conflicts, or changes in behaviour which are unexpected.

[Other]
SF# f00396266