IB/hfi1: Fix tstats alloc and dealloc

The tstats allocation is done in the accelerated ndo_init function but the
allocation is not tested to succeed.

The deallocation is not done in the accelerated ndo_uninit function.

Resolve issues by testing for an allocation failure and adding the
free_percpu in the uninit function.

Fixes: aa0616a9 ("IB/hfi1: switch to core handling of rx/tx byte/packet counters")
Link: https://lore.kernel.org/r/1642287756-182313-5-git-send-email-mike.marciniszyn@cornelisnetworks.com


Reviewed-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>