fix(starrocks): unwrap CLONE_EXPR and cast FE-narrowed builtins to their declared type

[aocsa]

Description

Two expression-translator fixes the multi-CN branch needed to run TPC-H. Both stand alone from the translator stack.

CLONE_EXPR. TPC-H q01's projection reads l_extendedprice twice, in two products. The FE's CloneDuplicateColRefRule wraps the second use in CLONE_EXPR because the BE has no copy-on-write. The translator had no arm for it and refused the whole fragment as an unsupported expression. The node has no value semantics, so the new translate_clone returns the child unchanged. I chose not to re-cast it to its declared type. The type is the child's by construction, and a cast would re-narrow a product the arithmetic path already lowered to FP64, so the cloned column would differ from the uncloned one the FE says it equals.

FE-narrowed builtins. year, month, day, length and char_length bind through DuckDB, where they return BIGINT. The FE declares SMALLINT, TINYINT, TINYINT, INT and INT. When the fragment's row crosses an exchange, the receiver declares its stream from the FE slots and the hop's schema guard refuses the BIGINT column. translate_function_call now wraps those five calls in a throwing cast back to the declared type through the new cast_to helper. The scalar function itself states the BIGINT the engine produces. The binder elides a cast to a type the expression already has, so the cast costs nothing where the two agree.

How I tested it. On a GB200 box (aarch64) I ran the CI trio: cargo fmt, clippy with warnings as errors, and the CN test suite without the engine feature. All 176 tests pass, 5 of them new.

Changes sit in experimental/starrocks/crates/starrocks-plan-translator, four files. type_mapper::i64_type is new. lib.rs gains one row in the expression table.

Five new tests in tests/translate.rs, among them clone_expr_is_transparent, fe_narrowed_functions_cast_to_declared_return_type and matching_return_type_function_stays_cast_free.

Not handled. Other FE-narrowed builtins are not in the match; add them as they show up. Aggregates and slot resolution are untouched.

Checklist

• Read CONTRIBUTING.md and ensure PR meets "reviewability" checklist
• Cover changes with new or existing tests
• Document configuration changes in code and summarize in the description above
• Update human and agent documentation (README.md, docs/, skills, CLAUDE.md)

References

• Refs fix(starrocks): lower decimal operations to fp64 #1236, the decimal to FP64 lowering. The CLONE_EXPR rule returns the child as is so it cannot undo that lowering.
• Carved from the multi-CN source branch aocsa/feat/pin-table-cn; independent of the translator stack (T1 to T4), which can merge in any order relative to this PR.