Speed up codespell:ignore check by skipping the regex in most cases

The changes to provide a public API had some performance related costs
of about 1% runtime. There is no trivial way to offset this any
further without undermining the API we are building. However, we can
pull performance-related shenanigans to compenstate for the cost
introduced.

The codespell codebase unsurprisingly spends a vast majority of its
runtime in various regex related code such as `search` and `finditer`.

The best way to optimize runtime spend in regexes is to not do a regex
in the first place, since the regex engine has a rather steep overhead
over regular string primitives (that is the cost of flexibility). If
the regex rarely matches and there is a very easy static substring
that can be used to rule out the match, then you can speed up the code
by using `substring in string` as a conditional to skip the
regex. This is assuming the regex is used enough for the performance
to matter.

An obvious choice here falls on the `codespell:ignore` regex, because
it has a very distinctive substring in the form of `codespell:ignore`,
which will rule out almost all lines that will not match.

With this little trick, runtime goes from ~5.6s to ~4.9s on the corpus
mentioned in #3419.

Author: nthykier

codespell_lib/spellchecker.py

@@ -109,7 +109,10 @@
 
 _builtin_default_as_tuple = tuple(_builtin_default.split(","))
 
-_inline_ignore_regex = re.compile(r"[^\w\s]\s?codespell:ignore\b(\s+(?P<words>[\w,]*))?")
+_codespell_ignore_tag = "codespell:ignore"
+_inline_ignore_regex = re.compile(
+    rf"[^\w\s]\s?{_codespell_ignore_tag}\b(\s+(?P<words>[\w,]*))?"
+)
 
 
 class UnknownBuiltinDictionaryError(ValueError):
@@ -210,6 +213,8 @@ def __init__(
             self.load_builtin_dictionaries(builtin_dictionaries)
 
     def _parse_inline_ignore(self, line: str) -> Optional[FrozenSet[str]]:
+        if _codespell_ignore_tag not in line:
+            return frozenset()
         inline_ignore_match = _inline_ignore_regex.search(line)
         if inline_ignore_match:
             words = frozenset(