test: add SQL file tests for try_to_date, try_to_timestamp, try_make_timestamp

[andygrove]

Which issue does this PR close?

N/A. This adds test coverage only. Surfaces and references #4554.

Rationale for this change

The try_* datetime parsing/constructor functions (try_to_date, try_to_timestamp, try_make_timestamp) had no Comet test coverage. They are RuntimeReplaceable and lower to Cast / GetTimestamp / MakeTimestamp, whose non-try forms Comet already accelerates, so it was worth checking empirically whether the try_ forms are accelerated too. The result is mixed and worth pinning down with tests.

What changes are included in this PR?

Adds spark/src/test/resources/sql-tests/expressions/datetime/try_datetime.sql, covering column and literal arguments (including invalid inputs that exercise the NULL-on-error path) and NULLs. Empirical findings, encoded in the query modes:

• try_to_date and try_to_timestamp currently fall back to Spark, so their queries use query spark_answer_only (result correctness is still validated against Spark).
• try_make_timestamp runs natively for valid inputs, but returns wrong values instead of NULL for out-of-range components (failOnError = false is not honoured natively). Valid-input queries use query spark_answer_only; invalid-input queries use query ignore(...) referencing the bug, try_make_timestamp returns wrong values for invalid inputs instead of NULL #4554.

When #4554 is fixed and the try_to_* fallback is addressed, these queries can be promoted to the default query mode to assert native execution.

How are these changes tested?

This PR is test-only. The file runs under CometSqlFileTestSuite, executing each query through both Spark and Comet and comparing results. Verified locally:

./mvnw test -Dsuites="org.apache.comet.CometSqlFileTestSuite try_datetime" -Dtest=none

Result: the discovered test file passes.