…result types

The `if` branch of `IfTransformStringsToEnumPass` was missing a check on the
function result type before attempting the string-to-enum optimization. When
`group_by_use_nulls = true` is combined with `GROUP BY ... WITH CUBE`, the
`if` function's result type becomes `Nullable(String)` instead of `String`.
The optimization then wraps into `toString`, which returns `Nullable(String)`,
triggering the debug assertion `isString(function_node.getResultType())`.

The `transform` branch already had this guard (line 157). Added the same
`isString(function_node->getResultType())` check to the `if` branch.

https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=96712&sha=ebc729143128f68f2f99c7a88691dc085c4574d0&name_0=PR&name_1=AST%20fuzzer%20%28amd_debug%29

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The previous fix (adding `isString` check on the `if` branch) caused a
mismatch between GROUP BY key and projection expressions when
`group_by_use_nulls` is used with CUBE/ROLLUP. The analyzer clones GROUP
BY keys for the projection and calls `convertToNullable` on the clones,
setting `wrap_with_nullable = true`. The `isString` check would skip the
optimization for the projection node (Nullable(String)) but apply it to
the GROUP BY key (String), creating a structural mismatch that causes
NOT_FOUND_COLUMN_IN_BLOCK errors.

The correct fix: allow the optimization to run on both nodes consistently,
and fix the assertion in `wrapIntoToString` to accept `Nullable(String)`
via `removeNullable`. Also fix the `transform` branch's `isString` check
similarly.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>