Load exceptions last in Tokenizer.from_bytes (explosion#12553)

In `Tokenizer.from_bytes`, the exceptions should be loaded last so that
they are only processed once as part of loading the model.

The exceptions are tokenized as phrase matcher patterns in the
background and the internal tokenization needs to be synced with all the
remaining tokenizer settings. If the exceptions are not loaded last,
there are speed regressions for `Tokenizer.from_bytes/disk` vs.
`Tokenizer.add_special_case` as the caches are reloaded more than
necessary during deserialization.

spacy/tokenizer.pyx

@@ -834,10 +834,12 @@ cdef class Tokenizer:
  self.token_match = re.compile(data["token_match"]).match
  if "url_match" in data and isinstance(data["url_match"], str):
  self.url_match = re.compile(data["url_match"]).match
- if "rules" in data and isinstance(data["rules"], dict):
- self.rules = data["rules"]
  if "faster_heuristics" in data:
  self.faster_heuristics = data["faster_heuristics"]
+ # always load rules last so that all other settings are set before the
+ # internal tokenization for the phrase matcher
+ if "rules" in data and isinstance(data["rules"], dict):
+ self.rules = data["rules"]
  return self