BUG: Handle inline images containing EI sequences

[stefan6419846]

Closes #3107.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.52%. Comparing base (038662a) to head (8253fdf).
Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3152      +/-   ##
==========================================
+ Coverage   96.51%   96.52%   +0.01%     
==========================================
  Files          53       53              
  Lines        8875     8903      +28     
  Branches     1622     1631       +9     
==========================================
+ Hits         8566     8594      +28     
  Misses        185      185              
  Partials      124      124              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[j-t-1]

Nice.

is_followed_by_binary_data(stream: IO[bytes], length: int = 10)

Remove spaces around =?

[j-t-1]

if byte < 32 and byte not in WHITESPACES_AS_BYTES:
            # This covers all characters not being displayable directly.

Suggestion for amending the comment:
C0 control codes except those defined as PDF whitespace characters.

Are the C1 control codes needed in the if statement?

[stefan6419846]

Remove spaces around =?

This will fail linting (ruff) as the whitespace is indeed correct here.

C0 control codes except those defined as PDF whitespace characters.

Are the C1 control codes needed in the if statement?

Yes, I might update the comment. But I am having some troubles understanding what you mean by C0 and C1 control codes - what are you referring to?

[j-t-1]

https://en.m.wikipedia.org/wiki/Unicode_control_characters

[stefan6419846]

Thanks for the link. I am not completely sure about this, but as the value 32_10 = 0x20 is the same as used by iText, this seems like the easiest approach for now. If we have an actual example where the C1 control codes are relevant, we might reconsider this in the future, but for now, this PR should fix this rare edge case in most conditions due to the additional operator guessing logic.