Preserve and validate native ARRAY INSERT VALUES

[osipovartem]

Summary

• Plan native ARRAY INSERT VALUES using the target column type without coercing unrelated VALUES columns. Preserve batch conversion for homogeneous rows and use per-row conversion only when mixed origins require it.
• Reject unsupported scalar subqueries in ARRAY VALUES at planning time instead of producing a plan that fails during execution.
• Validate legacy Utf8-backed ARRAY roots when the destination is native ARRAY; leave ordinary VARIANT targets and native List inputs on their existing conversion paths.

Verification

• cargo +1.97.0 test -p datafusion-sql --test sql_integration: 621/621 passed.
• Rustice at this commit: 38 focused ARRAY/VARIANT tests passed, legacy text-root execution test passed, and CSV/JSON COPY tests passed.
• Independent read-only review: approved with no blocking correctness, performance, or API findings.

Performance

The new strict conversion applies only to textual sources targeting native ARRAY. Homogeneous ARRAY VALUES are converted as a batch in the outer INSERT projection. A focused 1M-row conversion microbenchmark measured ~2.76M rows/s; end-to-end interactive latency has not been remeasured here.