Continue to break datafusion crate into smaller crates

**Is your feature request related to a problem or challenge? Please describe what you are trying to do.**
Quoting @Jimexist on https://github.com/apache/arrow-datafusion/issues/1750

> It helps with code management and dependency reasoning

Also I have heard annecdotal reports that compiling datafusion takes too long / too much memory to compile which is sad.

**Describe the solution you'd like**
- [ ] #5405
- [ ] Split out file_formats (maybe parquet gets its own?)
- [x] https://github.com/apache/arrow-datafusion/issues/1754

**Describe alternatives you've considered**
Leave as is

**Additional context**
First part of split was in https://github.com/apache/arrow-datafusion/issues/1750, driven by @Jimexist ❤️

Start by reviewing the referenced issue #5405 and the prior split in issue #1754, then inspect how the remaining file_formats code is organized. Done means agreeing on and implementing the remaining crate boundaries, including whether Parquet should have its own crate, while addressing the stated code-management and compilation concerns.

Written by the indexing model from the issue text.