Writers (DaliSpark, Spark SQL, Trino DML, Java Iceberg API) may submit
column names with different casing than what the table stores (e.g.
"id" vs "ID"). Because validateWriteSchema is case-sensitive, these
writes were rejected with InvalidSchemaEvolutionException even when the
field IDs matched, making case-insensitive writes impossible without
changing the table's existing column names.

Fix: in doUpdateSchemaIfNeeded, normalize the write schema's top-level
column names to match the table schema's casing (matched by Iceberg
field ID, not by name) before any comparison or storage. The table's
existing casing is never mutated, and writers do not need to know or
match the exact casing stored in the catalog.

Tables where two or more top-level columns share the same case-folded
name (e.g. "id" and "ID") are excluded from normalization because the
target column would be ambiguous, so writes to such tables must still
use exact casing.

Testing:
- BaseIcebergSchemaValidatorTest: 10 unit tests covering
  hasCaseDuplicateFields and normalizeSchemaCasingToTable in isolation,
  including new-column passthrough and field attribute preservation.
- RepositoryTest (H2 Spring integration): 2 new tests exercising the
  full save() path through the Spring context:
  - testCaseInsensitiveWrite_succeeds_andPreservesTableCasing: creates
    a table with "ID", saves an update with "id", asserts no exception
    and that the stored schema still shows "ID".
  - testCaseInsensitiveWrite_blockedForCaseDuplicateTable: creates a
    table with case-duplicate columns, asserts that a mismatched-casing
    save still throws InvalidSchemaEvolutionException.

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

…isit

Replace the top-level-only loop implementations of hasCaseDuplicateFields
and normalizeSchemaCasingToTable with TypeUtil.SchemaVisitor-based
implementations that recurse through structs, list elements, and map
key/value types at any depth.

Key changes:
- hasCaseDuplicateFields: uses TypeUtil.visit to detect case-duplicate
  sibling fields at any nesting level; correctly treats same-named fields
  in different structs as non-duplicates; adds Locale.ROOT to toLowerCase
- normalizeSchemaCasingToTable: uses TypeUtil.indexById to build a single
  flat Map<Integer, NestedField> for the entire table schema in one O(n)
  walk, then TypeUtil.visit to rewrite field names at any depth; uses
  reference-equality short-circuit to avoid allocation for unchanged fields
- Both methods use Iceberg's visitor contract so new Type variants cause a
  compile error rather than a silent passthrough

Add tests covering nested struct, list element, and map value normalization
as well as the hasCaseDuplicateFields depth cases.

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