Add SchemaInferenceOptions options to infer_schema and option to configure int96 inference

[jaychia]

This PR addresses part 2 of #1527

It solves the problem of configuring arrow2's Parquet schema inference to infer Timestamp fields from Parquet Int96 fields differently based on user input.

1. Adds a new SchemaInferenceOptions struct which allows for configurability of how schema inference on Parquet files
2. Adds a int96_coerce_to_timeunit flag to configure how Parquet int96 fields are inferred as arrow Timestamps
3. Adds *_with_options variants of the infer_schema and parquet_to_arrow_schema APIs to take in the options

[codecov]

Codecov Report

Patch coverage: 95.23% and project coverage change: +0.02% 🎉

Comparison is base (87ab844) 83.02% compared to head (4e4279f) 83.05%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1533      +/-   ##
==========================================
+ Coverage   83.02%   83.05%   +0.02%     
==========================================
  Files         391      391              
  Lines       42786    42866      +80     
==========================================
+ Hits        35523    35602      +79     
- Misses       7263     7264       +1     

Files Changed	Coverage Δ
src/io/parquet/read/schema/convert.rs	94.68% <94.62%> (+0.47%)	⬆️
src/io/parquet/read/schema/mod.rs	100.00% <100.00%> (ø)

... and 5 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[jaychia]

Hi @ritchie46 and @sundy-li

Here's a follow-up PR to #1532

(see PR description for more details)

[sundy-li]

BTW, int96 seems to be deprecated in parquet, it's not a stable feature. https://issues.apache.org/jira/browse/PARQUET-323

[jaychia]

BTW, int96 seems to be deprecated in parquet, it's not a stable feature. https://issues.apache.org/jira/browse/PARQUET-323

Indeed, but it is still widely used and supported by many systems for backwards-compatibility reasons

Unfortunately because Parquet is a long-lived format, and many enterprises use old versions of data frameworks, these deprecated features tend to live long after their deprecation :)