Runtime/Compiler - Support ComputeConfig::unpack_to_dest_mode #879

[kmabeeTT]

Note: This is blocked on tt-metal uplift. Will rebase + change target to main when tt-metal uplift is merged. For ease of viewing I picked base as my own branch for now on this PR.

Closes #879

Relatively simple change to accommodate API change in tt-metal introduced at tenstorrent/tt-metal#12794

• Pass field (per CB) down from compiler to runtime to metal API
• Hardcoded to UnpackToDestMode::Default (preserve previous/default behavior) and for all 32 CBs for now because tt-metal requires this (will be relaxed soon), replace with compiler logic when ready.

==

About workaround:

1. Talked to Radomir who added constraint in tt-metal (field must be set for all 32 CBs) originally about relaxing it ( [Bug Report] Remove restriction that UnpackToDestFormat must be specified for NUM_CIRCULAR_BUFFERS=32 CBs tt-metal#14066 )
2. Not sure what proper change in TTIRToTTMetal.cpp (to set field for only in-use CBs) will look like, it's a compiler task to figure out once above metal fix is pulled in.

[kmabeeTT]

Cleaned up changes. Had discussion with @nsmithtt offline this morning, decided to:

1. Don't need to hardcode using 32, instead we already have a tablegen-generated code to get max value of enum, use that instead
2. Add assert that our value of 32 (from say 1+ ttkernel::getMaxEnumValForCBPort()) is equal to tt-metal's NUM_CIRCULAR_BUFFERS=32)
3. To do that, have tt-metal export that value (required in a tt-metal ticket already) and put it in our system-desc, and consume it in TTIRToTTMetal.cpp

Made changes for number-1 above already on PR, going to try to see what changes look like for 2,3 (with system-desc hardcoding num_circlular_buffers=32 until metal exports it).