[Bench] IO Encode Benchmark

[CaiusDai]

What I have done in this PR

• Benchmarks for File IO, Encoding & Decoding written in C++

Note

This file largely referenced existing benchmarks in arrow-parquet. Two major sources are arrow/cpp/src/parquet/encoding_benchmarks.cc and arrow/cpp/src/parquet/column_io_benchmarks.cc.
The reasons I write a new file instead of using the original benchmarks are:

1. To support ratio comparison, input data need to be the same. We need a general method to generate data. (Original benchmarks generate data using different methods)
2. Original column io benchmark only tested cases for Int64 Plain encoding.
3. We can now run benchmark in one go.

Warning

1. In order to run the benchmark, we need to either write a new CMakeLists or incorporate this file in arrow source code, the correct folder to place it is thirdparty/arrow/cpp/src/parquet/. Also, thirdparty/arrow/cpp/src/parquet/CMakeLists.txt needs to add this benchmark file by adding the following command at line 433.

add_parquet_benchmark(io_encode_benchmark SOURCES io_encode_benchmark.cc
                      benchmark_util.cc)

2. Current build_third_party.sh disabled the parquet benchmark generation. On my local machine, I successfully build the benchmarks by executing the following code in arrow/cpp/build.

cmake .. -DARROW_PARQUET=ON \
      -DARROW_OPTIONAL_INSTALL=ON \
      -DARROW_BUILD_BENCHMARKS=ON
make parquet-benchmarks 

In docker environment, I tried to use same command to build benchmarks using ninja parquet-benchmarks but the process will fail in linking stage.

Relating Issues:

#6
#7