Make Tokenizer methods const (#4014)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/4014

The Tokenizer class supports encode() and decode() functions. They (and all sub functions) have to be 'const' in order to guarantee the Tokenizer object itself isn't mutated during those calls, which means it can be used in a multi threaded environment and is stateless, e.g. 2 decoding runs (parallel or sequential) have no influence on each other.

Reviewed By: larryliu0820

Differential Revision: D58780400

fbshipit-source-id: 2265a7c5639af600a3ae37e1cee4c7c77ba781ee