Add reader properties

[sadikovi]

This PR adds ReaderProperties struct, similar to WriterProperties used for writing. Currently this struct only contains batch_size, which is used to be hard-coded.

Following is done:

• Added ReaderProperties into src/file/properties.rs
• Updated SerializedFileReader::new() method to take additional ReaderProperties
• Updated SerializedRowGroupReader::new() method to take additional ReaderProperties
• Updated FileReader and RowGroupReader traits to have properties() method that returns reference-counted reader properties
• Updated row iterator to take tree builder and pass batch size from properties

This is a fairly big change and has implications of incompatibility of API with prior versions.

• Pros: having configurable batch size and potentially other reader properties.
• Cons: API change, people do not change it themselves, it might be enough to just increase default batch size to 1024/4096.

[sadikovi]

@sunchao Could you review this PR and let me know if we should keep the change or postpone it for now? This is one of the (optional?) items on write-support issue. Thanks!

[coveralls]

Pull Request Test Coverage Report for Build 620

• 0 of 0 changed or added relevant lines in 0 files are covered.
• 147 unchanged lines in 5 files lost coverage.
• Overall coverage increased (+0.04%) to 95.589%

---

Files with Coverage Reduction	New Missed Lines	%
file/writer.rs	7	95.93%
file/reader.rs	14	97.08%
schema/printer.rs	20	70.52%
file/properties.rs	23	92.56%
record/reader.rs	83	88.61%
<!--	Total:	147		-->

Totals
Change from base Build 619:	0.04%
Covered Lines:	12460
Relevant Lines:	13035

---

💛 - Coveralls

[sunchao]

Thanks @sadikovi ! Just curious: do you have in mind what other reader properties we will add in future? I checked parquet-cpp/parquet-mr and didn't find many properties that people need to tune on the reader side.

[sadikovi]

Actually, not. The only setting that we have and that one is used in record reader (which could be deprecated in the future) is batch size.

What I can do is simply bumping up the batch size to 1024/4096 instead of introducing the whole reader properties and changing API.

Let me know what you think.

[sunchao]

What I can do is simply bumping up the batch size to 1024/4096 instead of introducing the whole reader properties and changing API.

Yes I'd prefer this approach. Thanks!