Support tt-metal semaphores

[pjanevskiTT]

Fix #915

Add support for all tt-metal semaphore functions, as well as some utility functions that are useful for using semaphores through compiler

For example, code below

Value zero = i32(0, builder);
auto sem_id = builder.create<ttkernel::GetArgValOp>(arithOrMathOp.getLoc(), zero);
auto sem_addr = builder.create<ttkernel::GetSemaphoreOp>(arithOrMathOp.getLoc(), sem_id);
auto sem_l1_ptr = builder.create<ttkernel::CastToL1PtrOp>(arithOrMathOp.getLoc(), sem_addr);
builder.create<ttkernel::NocSemaphoreWaitOp>(arithOrMathOp.getLoc(), sem_l1_ptr, zero);

will generate kernel code

int32_t v5 = get_arg_val<uint32_t>(v4);
int32_t v6 = get_semaphore(v5);
volatile tt_l1_ptr uint32_t*  v7 = reinterpret_cast<volatile tt_l1_ptr uint32_t*>(v6);
noc_semaphore_wait(v7, v4);

TODO

• Implement conversion to tt_l1_ptr properly

[nsmithtt]

This looks great! Minor comments inline

[rpavlovicTT]

I have addded similar function for reduce template args, I guess you didn't rebase yet.

[rpavlovicTT]

Nice changes, few small comments, but overall looks good.

[jdesousa-TT]

@nsmithtt / @xanderchin with regard to #915, please re-review. The existing comments have been addressed, and this branch rebased. Ready for merge.

[nsmithtt]

Looks great, thanks @jdesousa-TT for revitalizing this one.