Add case/accent-sensitivity option for column collation in the Schema API

Problem/Motivation

#3456964 and #3507777 Usernames no longer accent-insensitive, and #3419816 Add support for case-sensitive validation all trace back to the same gap: Drupal's PHP-side uniqueness validation (UniqueFieldValueValidator) assumes PHP string equality matches database equality, but MySQL/MariaDB's default collation (utf8mb4_general_ci) is both case- and accent-insensitive while PHP string comparison is neither. A prior fix, #3415582 Error registering a user with different case, patched the PHP-side comparison for case, but the same class of bug recurs for accents and will recur for any other collation-driven equivalence.

#1144644 Specify collation for a database table proposes a general, arbitrary collation-string API for schema columns, but has stalled for over a decade, partly on cross-database portability concerns — MySQL, PostgreSQL, and SQLite each name and support collations differently — and on index/join performance concerns raised in that issue's own discussion.

#1518506 Normalize case sensitivity across database engines explores the related problem of making case-sensitivity behavior uniform across drivers, but its most recent discussion argues against forcing uniform behavior and toward letting call sites opt in explicitly — the direction this issue takes.

Steps to reproduce

N/A — this is a feature proposal, not a bug report. See the attached PDF for a worked mechanism diagram and per-driver DDL comparison.

Proposed resolution

Add an optional, closed-vocabulary schema key — e.g. 'collation' => 'case_sensitive' — on char/varchar/text column definitions, generalizing the pattern already shipped in #1237252 (DB case sensitivity: allow BINARY attribute in MySQL). Each database driver maps the semantic value to its own concrete implementation:

• MySQL/MariaDB: COLLATE utf8mb4_bin
• PostgreSQL: COLLATE "C"
• SQLite: COLLATE BINARY

PostgreSQL and SQLite already default to case- and accent-sensitive comparison, so the flag is a no-op on those two drivers and only changes generated DDL on MySQL/MariaDB — the actual source of #3456964 (equality mismatch between PHP and database) and #3507777 (accent-insensitivity regression).

Remaining tasks

• Confirm scope and naming of the enum value(s) with maintainers (starting with case_sensitive; open question whether accent_sensitive needs to be a distinct value or is implied)
• Implement per-driver mapping in each driver's Schema.php (MySQL, PostgreSQL, SQLite)
• Test coverage across all three core-supported drivers
• Documentation for contrib/custom module authors

User interface changes

None.

Introduced terminology

• collation (schema key): an optional key on char/varchar/text column definitions.
• case_sensitive: the initial enum value, mapped per driver to a byte-sensitive collation.

API changes

New optional schema key, additive and backward-compatible. Omitting the key preserves each driver's current default behavior exactly; no existing schema definition is affected, and altering an existing column's collation still requires an explicit hook_update_N().

Disclosure

This IS was written by AI (surprise!) -- I had intended to rewrite it once a PoC was finished.