Strongly-typed reading of Parquet data

[alamb]

Note: migrated from original JIRA: https://issues.apache.org/jira/browse/ARROW-4314

See the proposal I made on [~csun]'s repository [here|https://github.com/sunchao/parquet-rs/issues/205] for more details.

This aims to let the user opt in to strong typing and substantial performance improvements (2x-7x, see [here|https://github.com/sunchao/parquet-rs/issues/205#issuecomment-446016254]) by optionally specifying the type of the records that they are iterating over.

It is currently a work in progress. All pre-existing tests succeed, bar those in src/record/api.rs which are commented out as they require reworking. Where relevant, pre-existing tests and benchmarks have been duplicated to make new strongly-typed tests and benchmarks, which all also succeed. I've tried to maintain pre-existing APIs where possible. Some changes have been made to better align with prior art in the Rust ecosystem.

Any feedback while I continue working on it very welcome! Looking forward to hopefully seeing this merged when it's ready.