Suspected bug related to read with Mutation

[hucome]

Bug description

[Expected behavior] and [actual behavior].

1. Scenario example

The data of a table is stored in a large parquet file, which has many rows, such as 20000 rows.
TableScan of a query scans the data file and filter the deleted rows refered to the given corresponding mutation information. The deleted rows of the mutation are random and range from 0 to 20000.
The readBatchSize_ of the TableScan operator is less than 20000, such as 10000, so the rows in the file will be fetched multiple times

2. Behavior

The result is that only the data result of the first read correctly filter the deleted rows, but the subsequent reading results are mostly not correct.

3. Test example

duplicate the data in tpch table nation, and delete some data of it, then do the following query

-- select n_nationkey, count(1) from nation group by nation;

correct result:

default velox incorrect result:

System information

Velox System Info v0.0.2
Commit: de2f016
CMake Version: 3.22.2
System: Linux-5.4.119-1-tlinux4-0008
Arch: x86_64
C++ Compiler: /usr/lib64/ccache/c++
C++ Compiler Version: 10.3.0
C Compiler: /usr/lib64/ccache/cc
C Compiler Version: 10.3.0
CMake Prefix Path: /usr/local;/usr;/;/data/code/impala/Impala42/toolchain/toolchain-packages-gcc10.4.0/cmake-3.22.2;/usr/local;/usr/X11R6;/usr/pkg;/opt

Relevant logs

No response

[yingsu00]

@hucome Is this an Iceberg read problem? Does it happen with DWRF or ORC files as well? If it's Iceberg, can you please try the fix in #10505?