Fix perplexity calculation and resulting overcompression

[cornzz]

@iofu728

For longllmlingua this is not a 100% correct fix, especially for short prompts with longllmlingua errors occur.
I hope one of the maintainers can give some feedback whether this is semi-correct

Related: #195, #61

See this comment for details: #61 (comment)

Before the change (top: Llama 2, bottom: GPT-2):

After the change:

Note that the undercompression for 250 and 500 tokens is caused by the second bug / design flaw described in this issue #196 under "Semi-related bug", Also, this was measured including the bugfix in #198 otherwise 50 and 100 tokens would not be compressed.

I did not run the tests as compression outputs will change and tests will fail

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Was this discussed/approved via a Github issue? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?