MDEV-38904: Fix latin7 collation corruption and my_convert infinite loop

[itzanway]

This PR addresses MDEV-38904, which caused false index corruption in latin7 tables and a subsequent server hang during string conversion. The fix is divided into two parts: correcting the collation logic and hardening the string conversion utility.

1. Collation Fix (strings/ctype-extra.c)
   The root cause was a transitivity violation in the latin7_general_ci collation. In the original sort_order_latin7_general_ci array, the hyphen (-, 0x2D) and the space ( , 0x20) were assigned weights that caused collisions during MyISAM/Aria index compression.

Changes: Adjusted the sort_order_latin7_general_ci weights to ensure that the space character (Index 32) is uniquely weighted as the minimum printable value and that the hyphen (Index 45) has a distinct weight.

Impact: This prevents CHECK TABLE from falsely reporting "Key in wrong position" and prevents the creation of circular B-tree pointers that caused the server to loop.

2. Failsafe Loop Fix (strings/ctype.c)
   Even in cases of table corruption, the server should not hang at 100% CPU. The previous implementation of my_convert_using_func and my_convert_fix did not explicitly force a pointer advancement when the character set's mb_wc function returned a length of 0 (encountered during malformed/corrupt byte sequence reads).

Changes: Added an explicit from++ advancement when cnvres == 0.

Impact: This ensures the conversion loop always terminates, replacing malformed bytes with a '?' placeholder instead of looping infinitely.

Bug- https://jira.mariadb.org/browse/MDEV-38904?filter=-4

[gkodinov]

Thank you for your contribution! This is a preliminary review.

Couple of things before I start reviewing the substance:

• Please squash all of your commits into a single one
• Please have a commit message that complies with CODING_STANDARDS.md
• Please set your text editor to not convert spaces to table or vice versa.
• Please do not do space only changes.
• Please add test cases.
• Please make sure all the buildbot hosts compile and run tests successfuly

After a brief consultation with the future final reviewer, I should add that we can't "adjust weights" on an existing collation. Especially if it's used to store data on disk. This is not backwards compatible. But this is just a heads up. To be resolved during the final review.