Add coerce_types flag to parquet ArrowWriter

[tustvold]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

As discussed in #1666 not all types can be represented within a parquet schema.

Describe the solution you'd like

The consensus appears to be to:

• By default faithfully round-trip the source data, performing no potentially lossy type conversion
• Add a coerce_types flag that will use the arrow cast kernels to coerce incompatible types prior to writing them

In particular

Date64

If not coerce_types, write as Int64 and embed logical type in arrow schema only. Otherwise case to Date32

Timestamp

If not coerce_types, write as is, setting LogicalType / ConvertedType only where appropriate.

If coerce_types, cast to a UTC timestamp with the closest supported time unit, likely needing #1936.

Interval

If not coerce_types, write as FixedSizeBinaryArray matching the arrow representation and store logical type in arrow schema.

If coerce_types, convert to the relevant parquet representation.

Describe alternatives you've considered

See #1666

[dsgibbons]

take

[alamb]

I think this issue was done in #6840

If that is not correct, please reopen / let me know what else needs to be done

[dsgibbons]

@alamb my original PR only handled Date64. I'm not sure if Interval and Timestamp are still outstanding.

[alamb]

reopening per @dsgibbons 's comments

[alamb]

@dsgibbons would you be willing to make a PR to complete Interval and Timestamp so we can close this issue?

[dsgibbons]

@alamb yes, but will probably take a while (1-2 months). If someone else wants to finish this off in the meantime then go ahead. I'll comment once I start working on this.

[alamb]

Thank you @dsgibbons !