Builtin Collective Communication Syntax

[nrmer]

• extend lark language

  • add mullstream
  • add routing description

• add mulstream type

  • patch the existing functions to support the changed lark file
  • add new irnodes
  • add call to optimizations

• add optimization pass for mulstreams

  • change data blocks
    • create communication pattern
      • support snake communication pattern
      • support grid communication pattern
      • support pipelined communication
      • refactor code to make it more readable and to use the same structure in all parts
    • non pipelined datablock creation
    • pipelined datablock creation (should already be implemented)
  • fix subgrid
    • non pipelined subgrid creation
    • pipelined subgrid creation
  • change compute blocks
    • add additional operations
      • summation
      • multiplication
    • non pipelined compute block changes
    • pipelined compute block changes for 1 element
    • support multiple elements in the pipelined case
    • make code more readable
    • support nested reduce statements

• code style changes

  • add additional comments to code
  • make code easier to read

• write documentation

  • document usage for the user
  • document how the code is built

• write a test suite

  • add new test files
  • support all previous tests

[luxfg]

Overall good WIP, see comments :)

I also added some minor comments that I know you are aware of regarding the syntax, just so we don't forget.