fix(drizzle-kit): preserve commas inside MySQL enum values in snapshots (#5957)

[HarperZ9]

Fixes #5957.

Problem

handleEnumType in drizzle-kit/src/serializer/mysqlSerializer.ts splits the enum body on every comma:

const values = str.split(',').map((v) => `'${escapeSingleQuotes(v.substring(1, v.length - 1))}'`);

MySqlEnumColumn.getSQLType() emits enum values as raw, single-quoted strings joined by commas, so a value that itself contains a comma gets split apart. For mysqlEnum('col', ['a,b', 'c']) (type enum('a,b','c')) the generated snapshot column type comes out corrupted as enum('','','c').

Fix

Take the content between the first ( and last ), strip the outer quotes, and split on the ',' delimiter between values rather than on every comma. Plain enums round-trip unchanged.

Test

Added add table with enum values containing commas in drizzle-kit/tests/mysql.test.ts, modeled on the existing add table with ts enum test (no database required). It asserts that a plain enum still produces enum('a','b','c') and a comma-containing enum produces enum('a,b','c'). The test fails on the current code (enum('','','c')) and passes with the fix.

Notes

• Scoped to the comma bug. Separately, getSQLType() emits values unescaped, so an enum value containing a literal ' produces malformed SQL; that looks like a pre-existing issue deserving its own PR and is not addressed here.
• I could not run the full pnpm test end to end locally because the workspace drizzle-orm dist build fails on Windows (a tsc path-separator issue in the build tooling, unrelated to this change). The regression test was run against the real generateMySqlSnapshot code path via vitest; CI should run the standard harness cleanly.

[HarperZ9]

Current verification note from a fresh Windows checkout:

• npx.cmd -y pnpm@10.6.3 --version -> 10.6.3
• npx.cmd -y pnpm@10.6.3 install --frozen-lockfile --force stops before Vitest with:

ERR_PNPM_FETCH_404 GET https://registry.npmjs.org/drizzle-kit/-/drizzle-kit-0.25.0-b1faa33.tgz: Not Found - 404

So I do not have a fresh local rerun of the targeted regression test from this checkout. This is an install/registry blocker before the test runner starts, not a failure of the enum-comma test. The branch itself is unchanged and still scoped to the serializer split bug plus the regression test described above.