multi-schema-change fast reorg has no chance to save checkpoint

[lance6716]

We use FlushModeAuto to save checkpoint

tidb/pkg/ddl/ingest/checkpoint.go

Line 199 in dded4ea

flushed, imported, _, err := TryFlushAllIndexes(s.flushCtrl, FlushModeAuto, s.indexIDs)

But when indexIDs has more than one elements, each index (engine) check flush based on shared last flush time and will update the time immediately

tidb/pkg/ddl/ingest/backend.go

Lines 287 to 288 in dded4ea

	shouldFlush = shouldImport ||
	time.Since(bc.timeOfLastFlush.Load()) >= interval

It's unlikely that all engines are flushed

tidb/pkg/ddl/ingest/flush.go

Lines 18 to 31 in dded4ea

	func TryFlushAllIndexes(flushCtrl FlushController, mode FlushMode, indexIDs []int64) (flushed, imported bool, failedIdxID int64, err error) {
	allFlushed := true
	allImported := true
	for _, idxID := range indexIDs {
	// TODO(lance6716): use flushCtrl.Flush(indexIDs, mode)
	flushed, imported, err := flushCtrl.Flush(idxID, mode)
	if err != nil {
	return false, false, idxID, err
	}
	allFlushed = allFlushed && flushed
	allImported = allImported && imported
	}
	return allFlushed, allImported, -1, nil
	}

so at caller checkpoint is not updated.