add local backend support for executing task_impl locally

[lambda7xx]

Description of changes:

• key idea: use local cuda allocator to allocate memory for OP and then execute the OP locally.
• Add local backend support for env.get_fwd_accessor to measure OP cost. Currently env.get_fwd_accessor only supports legion.
  • add ITaskArgumentAccessor, and LocalTaskArgumentAccessor and LegionTaskArgumentAccessor.
  • add cuda allocator
  • implement the LocalTaskArgumentAccessor

Related Issues:

Linked Issues:

• Issue #

Issues closed by this PR:

• Closes #

Before merging:

• Did you update the flexflow-third-party repo, if modifying any of the Cmake files, the build configs, or the submodules?

---

This change is 

[lockshaw]

Moved to #1097