Speed up spellchecking by ignoring whitespace-only lines

The new API has introduced extra overhead per line being spellchecked.
One way of optimizing out this overhead, is to spellcheck fewer lines.
An obvious choice here, is to optimize out empty and whitespace-only
lines, since they will not have any typos at all (on account of not
having any words).

A side-effect of this change is that we now spellcheck lines with
trailing whitespace stripped. Semantically, this gives the same result
(per "whitespace never has typos"). Performance-wise, it is faster in
theory because the strings are now shorter (since we were calling
`.rstrip()` anyway). In pratice, I am not sure we are going to find
any real corpus where the trailing whitespace is noteworthy from a
performance point of view.

On the performance corpus from #3491, this takes out ~0.4s of
runtime brining us down to slightly above the 5.6s that made the
baseline.

Author: nthykier

codespell_lib/_codespell.py

@@ -881,7 +881,8 @@ def parse_file(
     )
 
     for i, line in enumerate(lines):
-        if line.rstrip() in exclude_lines:
+        line = line.rstrip()
+        if not line or line in exclude_lines:
             continue
 
         extra_words_to_ignore = set()