Use Snakefiles instead of makefiles for benchmarks

[JosseVanDelm]

Update:

I got pretty much all kernels working with snakemake now, but there's still a lot of clean up to do:

• Remove all makefiles
• Deduplicate rules as much as possible
• Move TinyML perf to newer snakemake passes in CI
• clean up the altered datagen scripts, remove argparsers from datagen scripts

I'm going to list some annoying issues along the way:

• there are no rules yet for postprecessing and gathering results
• We can not run verilator in parallel:
  • --trace-prefix functionality has to be tested again
  • Also need --vcd-prefix functionality to be added for dense_matmul kernel
• Properly install snakemake in the docker container (we probably want to update python to 3.12 for this)
• Move to pandas dataframes for postprocessing in dense_matmul
• Deprecate snax_gemm, and snax_streamer_gemm in favor of snax_gemmx?

• Benchmarks
  • Tiled matmul ⚠️ this benchmark is in a poor state see Update accfg benchmarks #300
  • Dense matmul ✔️ Binary compilation, ⚠️ results postprocessing
• Kernels
  • mlperf tiny
  • rescale
  • alloc
  • gemmini
  • gemm
  • simple_copy
  • streamer_alu (binaries running sequentially)
  • streamer_matmul (binaries runnnig sequentially)
  • transform_copy (binaries running sequentially)
  • tiled_matmul (binaries running sequentially)

[JosseVanDelm]

Closing this, but keeping as reference