GeezerVison

Fixes #4324.

Generated non-null enums are defined string types. Passing them to mysqltsv.Encoder.AppendValue fails at runtime because they are neither a built-in string nor a driver.Valuer. The generated code compiles, but LOAD DATA cannot consume its rows.

Recognize the generated Go enum names, including the qualifier used for split models, and emit AppendString(string(value)). Nullable enum wrappers already implement driver.Valuer, so they keep the generic path. Type overrides are checked by their resolved Go type, not merely their SQL column type.

The production change is 11 added lines. The remaining files are CLI inputs and generated goldens for default arguments, parameter-struct pointers and split models. They cover multi-column and scalar enums, nullable values, an overridden enum, ordinary strings and INSERT/REPLACE conversion. Existing goldens, models, dependencies and public APIs are unchanged.

Validation:

• The final CLI golden fails on the clean base and passes in both base and managed-db contexts, also with -race.
• Real MySQL 9.7.2: the original generated code fails nine enum cases and passes twelve controls. All 21 cases pass after the fix, including exact stored values/counts and zero SHOW WARNINGS. Checked with -race on Go 1.26.8 and 1.27.0.
• Full tagged suite with the CI's Go 1.26.5, CGO_ENABLED=0, live PostgreSQL 16.14/MySQL 9.7.2 and all required process plugins: 6,889 passing test observations, zero failures. The 153 declared skips are unsupported replay contexts and one statement with no faithful rendering; none are missing tools.
• Full generated-code corpus builds. All six CI build targets (Linux/macOS/Windows, amd64/arm64) build with CGO disabled. Native build and vet pass on Go 1.26.8 and 1.27.0; generator race tests and formatting pass.
• govulncheck passes on Go 1.26.8. An additional scan on the older build/test compiler, Go 1.26.5, reports five standard-library advisories that also reproduce on the clean base. No toolchain/dependency changes are included.

REPLACE duplicate-key behavior stays with #4559. The runtime REPLACE cases use fresh keys to isolate the enum encoding problem; this PR does not claim to fix that separate semantic issue.