feat(expr): evaluate prefix predicates with strict metrics

[fallintoplace]

Which issue does this PR close?

N/A. Found while comparing StrictMetricsEvaluator with the Java reference implementation and the sibling metrics evaluators.

What changes are included in this PR?

Strict metrics evaluation answers whether every row in a data file must match a predicate. It previously returned false for startsWith and notStartsWith even when file bounds proved the result.

This PR adds conservative proofs for both predicates:

• startsWith returns true only when null metrics prove there are no nulls and both string bounds start with the requested prefix
• notStartsWith returns true when the column is entirely null, or when a truncated lower or upper bound proves the requested prefix is outside the possible value range
• missing or inconclusive metrics still return false
• non-string literals or bounds return an error instead of being evaluated as valid string metrics

The bound comparisons use iterators and do not allocate temporary truncated strings.

Are these changes tested?

Yes. Unit tests cover matching and non-matching bounds, missing bounds, null-containing files, all-null files, prefix overlap, and invalid non-string bounds.

Validated with:

• cargo test -p iceberg expr::visitors::strict_metrics_evaluator::test --lib
• cargo clippy -p iceberg --lib -- -D warnings