hash constant buffer to improve deduplication performance (#1185)

Summary:
Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ https://github.com/pytorch/executorch/issues/1185

Currently for each constant tensor we need to compare its buffer with
all the emitted buffer before to see if there's a duplicate. This is bad
especially if we have more than one ExecutionPlan in the program. This
PR introduces a map from the hash value to buffer index and hopefully
can largely reduce the deduplication time.

Pull Request resolved: https://github.com/pytorch/executorch/pull/1185

Reviewed By: tarun292, JacobSzwejbka

Differential Revision: D51182655

Pulled By: larryliu0820

fbshipit-source-id: df63bb800238c2021ae2355ad7d2e8cc13313067