[WIP] Guarantee temp table cleanup and fix its ordering vs schema update

[chikamura]

Summary

• Ruby's begin/ensure wraps everything from right after auto_create through the final schema patch, and always deletes the temp table (and cleans up local files) afterward, so cleanup happens even if the load, copy/merge, or schema update fails. The java port only wrapped the copy/merge step, deleted the temp table inline before the schema update, and skipped it if a delete came before an unrelated failure.
• Wrap the whole post-autoCreate body in try/finally so a failure anywhere (storeCachedSrcFieldsIfNeed, load, copy/merge, updateTableIfNeed) still guarantees the temp table delete, nest a second finally for the local intermediate file cleanup so a delete failure doesn't block it either, and move updateTableIfNeed() before the temp table delete so a delete failure no longer blocks the schema update from being attempted.
• Updated the existing MockWebServer-based tests to match the new call order (schema update now happens before the temp table delete), and added testRunReplaceModeStillDeletesTempTableWhenStoreCachedSrcFieldsIfNeedFails to cover the temp table delete being guaranteed even when storeCachedSrcFieldsIfNeed() itself fails.

Test plan

• ./gradlew test (105 tests, 1 skipped, all passing)
• ./gradlew spotlessApply

Note

Much of the diff is just re-indentation from the new try/finally nesting — reviewing with "Hide whitespace changes" enabled is recommended.

🤖 Generated with Claude Code

[chikamura]

Removed the mode switch since it was redundant: BigqueryTaskBuilder.setTempTable() only auto-populates temp_table for exactly these four modes, so task.getTempTable().isPresent() alone already implies the same restriction. This also matches ruby's own unconditional if task['temp_table'] check in its ensure block, with no mode restriction.

https://github.com/trocco-io/embulk-output-bigquery/blob/b4f8d2f4baf1c431a1cee3631790526441c3f9ec/lib/embulk/output/bigquery.rb#L498-L500