Add support for providing Chrono formats when parsing strings -> datetime/nanos and Parser implementations

[Omega359]

This is related to apache/datafusion#5398 where functions for parsing string -> datetime and string -> nanos would be added to include a second 'format' parameter. The format would be a chrono format (https://docs.rs/chrono/latest/chrono/format/strftime/index.html) and would return an Err if the string couldn't be parsed with the provided format.

The proposed two new functions that would be added would be :

• pub fn string_to_datetime_formatted<T: TimeZone>(timezone: &T, s: &str, format: &str) -> Result<DateTime<T>, ArrowError> (mirrors the existing string_to_datetime function)
• pub fn string_to_timestamp_nanos_formatted(s: &str, format: &str) -> Result<i64, ArrowError> (mirrors the existing string_to_timestamp_nanos function)

In addition, for completeness all the existing Parser implementations in that file would be updated to include a 'parse_formatted' implementation where missing.

[tustvold]

A few points:

• The current logic is quite highly optimised, we need to be careful not to introduce any performance regressions
• There are a couple of known bugs with the chrono format strings where the parser is overly permissive - %Y should return an error when parsing a non-4 digit year chronotope/chrono#332
• Timestamp parsing, especially timezones is very fiddly. I would request we keep code changes small, as it is very easy to subtly break things

Overall I am rather lukewarm on this functionality, as it is by definition diverging from established standards, but I will endeavour to review well tested and focused changes to this logic that don't regress existing workloads.

[Omega359]

Overall I am rather lukewarm on this functionality, as it is by definition diverging from established standards, but I will endeavour to review well tested and focused changes to this logic that don't regress existing workloads.

Would you prefer this to not be in arrow-cast and rather be left in datafusion? I'm only adding it here to mirror the existing functionality - I have no desire to change any of the current implementation details.

[tustvold]

Incubating functionality in DataFusion before lifting it into arrow-rs has been a successful model in the past, and might be a quicker way to iterate on apache/datafusion#5398 without having to wait on an arrow-rs releases.

[Omega359]

Sounds good, I'll pull in my changes to DataFusion for now and we can close this issue.

[Omega359]

Closing this issue with apache/datafusion#5398 as an alternative.