Speed up _is_eol_token

[correctmost]

Continuation of #1256 because I force pushed and cannot reopen that PR. (Sorry for the noise.)

If this PR seems too risky because of assumptions about tokenization, feel free to pass!

Stats

Before

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
  1478156    0.885    0.000    1.197    0.000 pycodestyle.py:1831(_is_eol_token)
  1472360    0.364    0.000    0.364    0.000 {method 'lstrip' of 'str' objects}

Command	Mean [s]	Min [s]	Max [s]	Relative
pycodestyle .	18.472 ± 0.196	18.067	18.848	1.00

After

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
  1478156    0.511    0.000    0.511    0.000 pycodestyle.py:1831(_is_eol_token)
   225341    0.055    0.000    0.055    0.000 {method 'lstrip' of 'str' objects}

Command	Mean [s]	Min [s]	Max [s]	Relative
pycodestyle .	18.360 ± 0.186	18.159	18.781	1.00

[correctmost]

Versions tested locally (on Arch Linux with pyenv):

• 3.13.0b4
• 3.12.4
• 3.12.0
• 3.11.9
• 3.11.0
• 3.10.14
• 3.10.0
• 3.9.19
• 3.9.0
• 3.8.19
• 3.8.0