Fix Markdown formatting offsets for non-ASCII URLs

[michaelsam94]

Summary

• count TextUrl URL arguments in UTF-16 code units when adjusting generated Markdown entity offsets
• add a regression for a non-ASCII URL followed by a preserved custom emoji entity

Fixes #3686.

Tests

• git diff --check upstream/master...HEAD
• Not run: CMake test target, because cmake is not installed in this environment.

[levlam]

The fix is wrong and changes nothing. get_markdown_v3 is never called with invalid UTF-8. I have already identified actual cause of the mentioned issue and fixed it in the private TDLib version, which is supposed to be publicly available soon.

[michaelsam94]

You are right; the invalid UTF-8 guard was the wrong fix. I force-pushed a corrected patch.

The issue I found is in get_markdown_v3: when closing a TextUrl, it added entity->argument.size() to utf16_added, but that argument can contain non-ASCII UTF-8. This can shift preserved entities after the link to invalid UTF-16 offsets in the generated Markdown before it is reparsed. The patch now uses text_length(entity->argument) and adds a regression with a Cyrillic URL followed by a custom emoji entity.

[levlam]

The patch is correct now, but the same fix has already been done in the private TDLib version.