fix: hybrid storage encode bug in multi record batch

[chunshao90]

Which issue does this PR close?

Closes #403

Rationale for this change

Fix bugs described in #403 .
Hybrid storage compresses multi-record batches into one record batch.
There is a bug where collapsible_col_arrays in TsidBatch only stores the first record batch data, other record batch data are lost.

What changes are included in this PR?

Are there any user-facing changes?

No.

How does this change test

Modify some tests to cover this bug.

[jiacai2050]

@chunshao90 I think of one critical issue for hybrid format: it will not be compatible with bloom filter.

We built bloom filter for each row group, there will be exact num_rows_per_row_group rows in one row group, since ParquetWriter will flush row group until the accumulated rows reach num_rows_per_row_group, otherwise it will just pending them.

The problem is that hybrid storage will shorten rows in one row group, for example if we set num_rows_per_row_group to 8192, after hybrid conversion, it will less than 8192, then ParquetWriter will not flush this row group since 8192 is not reached.

[chunshao90]

@chunshao90 I think of one critical issue for hybrid format: it will not be compatible with bloom filter.

We built bloom filter for each row group, there will be exact num_rows_per_row_group rows in one row group, since ParquetWriter will flush row group until the accumulated rows reach num_rows_per_row_group, otherwise it will just pending them.

The problem is that hybrid storage will shorten rows in one row group, for example if we set num_rows_per_row_group to 8192, after hybrid conversion, it will less than 8192, then ParquetWriter will not flush this row group since 8192 is not reached.

I create an issue #435.

[jiacai2050]

LGTM