Fix enum-enum math for compatibility with C++20

[r-barnes]

When compiling code that includes the header infiniband/verbs.h with CUDA using C++20 we observe:

buck-out/v2/gen/fbsource/third-party/rdma-core/stablev60/__ibverbs__/c01e142197f8fe04/buck-headers/infiniband/verbs.h:573:59: error: bitwise operation between different enumeration types ('(unnamed enum at buck-out/v2/gen/fbsource/third-party/rdma-core/stablev60/__ibverbs__/c01e142197f8fe04/buck-headers/infiniband/verbs.h:562:1)' and 'ibv_create_cq_wc_flags') is deprecated [-Werror,-Wdeprecated-anon-enum-enum-conversion]
  573 | IBV_CREATE_CQ_SUP_WC_FLAGS = ((((((IBV_WC_STANDARD_FLAGS) | (IBV_WC_EX_WITH_COMPLETION_TIMESTAMP)) | (IBV_WC_EX_WITH_CVLAN)) | (IBV_WC_EX_WITH_FLOW_TAG)) | (IBV_WC_EX_WITH_TM_INFO)) | (IBV_WC_EX_WITH_COMPLETION_TIMESTAMP_WALLCLOCK))
      |                                   ~~~~~~~~~~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This use of enums seems to be a pre-C++11 way of generating compile-time constants. This PR uses explicit casts to suppress the compile-time warning.

This code, is of course C, but the enum-enum conversion issues leak out into C++ because they're in a header. There are a few options:

• Use explicit casts to suppress the warning (what I've done here)
• Change the code so it uses constants - the enums don't seem semantically meaningful here.
• Upgrade to the C23 standard to gain access to constexpr (heavyweight solution).
• Suppress the relevant compiler flags in this header using #pragma clang diagnostic push and #pragma clang diagnostic pop (it's more complicated than this, but folly has a macro that could be copied to cover all scenarios)

The current code's use of enums seems as though it could cause increasing compilation issues over time, so fixing it seems like a good idea.

Let me know what you'd like to do - I'm happy to help.

[jgunthorpe]

I think since these are intended to be used by an application that will | them together the only answer is to change them to #define's. Otherwise the app will hit these warnings too.

[jgunthorpe]

And if we are going to do this we should do it for all the enum's being used for defining bit fields.