Benchmarks, bug fixes, and writer improvements

Add benchmarks vs parquet and raw json data
Fixes #112, which caused lance datasets to be written with duplicated Arrow buffers
Adds support for large binary, boolean, and temporal types.

What's Changed

• Restore the custom build script for pyarrow by @changhiskhan in #84
• Add support for Arrow temporal types, large binarys, and null type by @eddyxu in #85
• Fix warnings for CMP0135 by @eddyxu in #86
• Remove the primary key for now by @changhiskhan in #88
• reproduce large binary write bug by @changhiskhan in #91
• Basic extension type handling by @changhiskhan in #90
• Make FileReader pass schema as shared_ptr to Project by @eddyxu in #100
• Fix compile warnings for testing by @eddyxu in #103
• Fix length calculation when reading list array with offset by @eddyxu in #104
• Add read/write support for time / datetime types by @eddyxu in #106
• Set batch size when writing the file by @eddyxu in #111
• Coco benchmarks for lance and parquet formats by @changhiskhan in #97
• Minor fixes for benchmarks by @changhiskhan in #115
• Project skips scan if the filter covers all the columns by @eddyxu in #117
• Rebind filter after lance schema projection. by @eddyxu in #118
• Plain encoding supports writing boolean array by @eddyxu in #120
• Update README.md by @changhiskhan in #123
• PlainEncoding uses raw value pointer to write data to disk. by @eddyxu in #116

Full Changelog: v0.0.3...v0.0.4