Parquet: Fix initial-default rows dropped when filtering on the defaulted column

[cbb330]

What

A column added by schema evolution with an initial-default is backfilled with the default at read
time, but ParquetMetricsRowGroupFilter evaluates predicates against a column that is absent from
a data file as if it were all-null. For a file written before the column existed, a predicate
referencing the column — including the IsNotNull engines infer for null-intolerant predicates —
skipped the row group, silently dropping exactly the rows the default backfills. The full scan
was correct, so the case was untested and unobserved.

Fixes #16690.

Repro (Spark 3.5, Parquet, format-version 3)

CREATE TABLE t (id bigint, name string) USING iceberg TBLPROPERTIES ('format-version'='3');
INSERT INTO t VALUES (1, 'Alice');                          -- written before column c exists
-- table.updateSchema().addColumn("c", StringType, lit("US")).commit();
INSERT INTO t VALUES (2, 'Bob', 'US');                      -- c present

Query	Before	After
SELECT id, c	(1,US),(2,US) ✅	(1,US),(2,US)
WHERE c = 'US'	(2) ❌	(1),(2) ✅
WHERE upper(c) = 'US'	(2) ❌	(1),(2) ✅
WHERE c IS NOT NULL	(2) ❌	(1),(2) ✅
WHERE c = 'CA'	()	() ✅ (absent file still excluded)
WHERE c IS NULL	()	() ✅

Fix

The drop happens in ParquetMetricsRowGroupFilter, whose per-predicate handlers treat a column
missing from the file as all-null (// the column is not present and is all nulls). The filter is
already constructed with the table read schema, which carries initialDefault, so it has everything
it needs to be correct.

Override the single predicate dispatch point: when a predicate references a column that is missing
from the file but has an initialDefault, evaluate it against the default value
(pred.test(default)) instead of falling through to the null-assuming handlers. c = 'US' →
ROWS_MIGHT_MATCH, c = 'CA' → ROWS_CANNOT_MATCH, IsNotNull(c) → match, IsNull(c) → skip.
Columns present in the file, and missing columns without a default, are unchanged — so present-column
files still prune on real stats, and existing missing-column behavior is preserved.

This mirrors how the filter already gets the schema; it's a fix at the source rather than a rewrite
of the predicate at the call site.

Testing

• TestMetricsRowGroupFilter#testColumnNotInFileWithInitialDefault (Parquet) — unit coverage for the
  fold: match/non-match/IsNull/IsNotNull/notEqual against the default.
• TestDefaultValuesFilteredRead (Spark 3.5) — end-to-end; fails on main without the fix.
• Full :iceberg-parquet:test, :iceberg-data:test TestMetricsRowGroupFilter (Parquet + ORC), and
  :iceberg-arrow:test pass.

Scope / notes

• ParquetReader and VectorizedParquetReader both share this filter via ReadConf, so the row and
  vectorized read paths are both fixed.
• The dictionary and bloom row-group filters do not skip a column missing from the file, so they need
  no change (verified by the end-to-end test).
• Avro applies no file-level filtering (residual is applied by the engine after the default is
  injected), so it is already correct. ORC throws UnsupportedOperationException on reading an
  initial-default column, so there is no silent drop there.
• The record-level filter path (ParquetFilters.convert) shares the same missing-column assumption
  but is only reached via the @Deprecated readSupport reader, not by any engine read; it can get
  the same treatment as a follow-up if desired.

[pvary]

Why is this not part of the ParquetFilters.convert method?