Refactor synthetic utilities and add multidevice support

[wlruys]

• Rename CUDA to GPU across the board, this will make HIP support more natural in the future than treating it as a separate device type.
• Changes task graph and data config export files to YAML instead of the old custom "parla graph" grammar. This is to make it more maintainable. Conversions to and from the old files are provided.
• Extends task and data configs for multidevice. Each device in an multidevice setup can have a different task config (different runtime/memory/etc)
• Changes launcher and busy_wait kernels to support multidevice
• Separated mapping from placement. Allow mapping files to be saved and loaded onto a task graph.
• Added a field for task ordering
• Updated serial & ind to write to the Task graph table instead of the file format.

This is still a work in progress. Remaining work items:

• Port all synthetic graphs to new generation style
• Add logging for multidevice mapping
• Refactor log parsing and add support for extracting the graph from the log
• Fix verification methods to prevent false positives on runahead. Extend them to all phases.
• Add crosspy support to launcher
• Add data movement verification
• (Optional) Move visualization methods from old repo to be part of this utility submodule
• (Optional, will be new PR) Move task graph simulator from old repo to be part of this utility model. Refactor so it's easy to use for someone that didn't write it 😅