Map renamed Iceberg columns by field ID

[osipovartem]

Summary

• resolve renamed Iceberg columns by their stable Parquet/Iceberg field ID instead of the physical file name
• preserve the logical field name, metadata, and nullability returned to DataFusion
• apply the same field-ID lookup to renamed Variant columns
• pre-index physical fields once per file-schema adapter

Performance

• unchanged names retain the existing DefaultPhysicalExprAdapter fast path
• same-type renames use a zero-copy expression that returns the original Arrow array
• type casts are created only when the physical and logical data types differ
• field-ID lookup is O(1) from a per-adapter map; there is no per-row conversion

Why

Iceberg column renames preserve field IDs and do not rewrite existing Parquet files. The previous adapter resolved ordinary columns by name, causing renamed non-null columns to fail as physically missing.

Validation

• cargo +1.95.0 test -p datafusion_iceberg --lib --locked (83/83)
• cargo +1.95.0 clippy -p datafusion_iceberg --lib --tests --locked -- -D warnings
• the focused test asserts Arc::ptr_eq between the physical and renamed output arrays