Try out a different IdKind table approach

[jonmeow]

I was thinking about these after #5526, was wondering how others will feel about this kind of approach:

• Adding a helper to TypeEnum to get the table construction.
• In what were previously Make functions, return the element instead of returning the table.
• By returning the element, no more need to pass in a nullptr (now have a concrete instance).

I think this is a mild simplification, but maybe worth it.

Note, would appreciate it if there are thoughts on how to provide a boilerplate Invalid implementation (maybe it'd be fine to just return nullptr and cause a crash that way, but I was hesitant to do that).

[jonmeow]

Note, I also tried e.g.:

    static constexpr std::array<AddFnT, SemIR::IdKind::NumValues> Table = {
        [](Worklist& worklist, int32_t arg) {
          worklist.Add(Inst::FromRaw<Types>(arg));
        }...,
        [](Worklist& /*worklist*/, int32_t /*arg*/) {
          CARBON_FATAL("Unexpected invalid argument kind");
        },
        /*none=*/[](Worklist& /*worklist*/, int32_t /*arg*/) {}};

But Clang doesn't like that:

llvm-built/bin/../include/c++/v1/array:191:15: error: '__elems_' declared as array of functions of type 'void (Carbon::SemIR::(anonymous namespace)::Worklist &, int)'
  191 |   _Tp __elems_[_Size];
      |               ^
toolchain/sem_ir/inst_fingerprinter.cpp:320:60: note: in instantiation of template class 'std::array<void (Carbon::SemIR::(anonymous namespace)::Worklist &, int), 41>' requested here
  320 |     static constexpr std::array<AddFnT, IdKind::NumValues> Table = {
      |                                                            ^
1 error generated.

I don't see how to fix this diagnostic. That's why I settled on the MakeTable helper, but maybe I'm missing some syntax detail.