Consolidate string compare throughout

[mhofman]

Describe the bug

We have different ways of comparing strings, none of which are "correct". Now that's we've updated to a version of endo with endojs/endo#2008, we can use its compareByCodePoints throughout.

To Reproduce

Incorrect string comparison

• a < b or any variation. This results in a comparison by UTF-16 code units.
• [a, b].sort() similarly uses code units
• a.localeCompare() compares by environment specific local rules. The SES shim is supposed to tame this by default, but I have seen surprising results (see Consistent store order #10299 test and undo localeCompare change).

Expected behavior

Comparison by codepoints which we're trying to standardize on, and is the behavior for any implementation that compares utf-8 encoded bytes like SQLite, golang (doesn't strictly requires that string be utf-8, but literals and range iterations are), and JSON (which specifies utf-8 as the default encoding).

Additional context

#10299 fixes the more egregious of these cases, which caused a discrepancy in the order of Stores when they're heap or virtual/durable, and whether they're in a real or fake liveslots environment. In comments, @gibson042 also points out at other cases of mistaken comparisons (likely not exhaustive).