[SPARK-56638][SQL] Eliminate legacy error class for TABLESAMPLE fraction validation

[XdithyX]

What changes were proposed in this pull request?

This PR replaces the legacy parser error class used by invalid TABLESAMPLE fractions with a named error class (SPARK-56638).

Specifically, this PR:

• Adds INVALID_TABLESAMPLE_FRACTION to error-conditions.json
• Adds QueryParsingErrors.invalidTableSampleFractionError
• Replaces the generic ParserUtils.validate(...) call in AstBuilder.withSample(...) for TABLESAMPLE fraction validation
• Updates PlanParserSuite to assert the new error class

The validation logic is unchanged. TABLESAMPLE fractions are still required to be in the [0, 1] interval, allowing the existing rounding epsilon.

Why are the changes needed?

ParserUtils.validate(...) always throws _LEGACY_ERROR_TEMP_0064, which is a generic legacy error class.

For invalid TABLESAMPLE fractions, the parser already knows the specific failure: the computed sampling fraction is outside the allowed [0, 1] interval. Using a named error class makes the error condition explicit and continues the ongoing cleanup of legacy temporary error classes.

This PR does not update ParserUtils.validate(...) itself because that helper is shared by unrelated parser validations. Making it throw INVALID_TABLESAMPLE_FRACTION would incorrectly label non-TABLESAMPLE parser failures as TABLESAMPLE fraction errors.

The new error uses SQLSTATE 22023 because the statement is syntactically valid, but the supplied numeric value is invalid/out of range. This is consistent with existing Spark error conditions for invalid argument or range values, such as INVALID_NUMERIC_LITERAL_RANGE and other 22023 value-validation errors, rather than 42601, which is used for syntax errors.

Does this PR introduce any user-facing change?

Yes, for invalid TABLESAMPLE fractions, the error class changes from _LEGACY_ERROR_TEMP_0064 to INVALID_TABLESAMPLE_FRACTION.

How was this patch tested?

Ran build/sbt 'catalyst / Test / testOnly org.apache.spark.sql.catalyst.parser.PlanParserSuite -- -z "sampled relations"'
Also regenerated and verified the affected SQL golden files:
SPARK_GENERATE_GOLDEN_FILES=1 build/sbt 'sql / Test / testOnly org.apache.spark.sql.SQLQueryTestSuite -- -z "tablesample-negative.sql"'
SPARK_GENERATE_GOLDEN_FILES=1 build/sbt 'sql / Test / testOnly org.apache.spark.sql.SQLQueryTestSuite -- -z "pipe-operators.sql"'
build/sbt 'sql / Test / testOnly org.apache.spark.sql.SQLQueryTestSuite -- -z "tablesample-negative.sql"' 'sql / Test / testOnly org.apache.spark.sql.SQLQueryTestSuite -- -z "pipe-operators.sql"'

Was this patch authored or co-authored using generative AI tooling?

Yes. Generated-by: OpenAI GPT-5.5 Codex

[XdithyX]

Hi @cloud-fan and @sarutak, could you please take a look? Thanks.
Tagging @HeartSaVioR as well.