Polygonize: Write parquet compliant to fiboa with additional metadata #53

[m-mohr]

Closes #53 and #26

inference download/run:

• Gets last timestamp from STAC Items, passes them through the TIFF files via TIFFTAG_DATETIME metadata.

inference polygonize:

• Creates GeoParquet 1.1 files compliant to fiboa using fiboa CLI code.
• Adds columns determination_datetime, determination_method, perimeter.
• area was generally changed from m² to ha.
• Other file formats are also "fiboa compliant" (with regards to the columns).

Drawback:

• All other non-Parquet formats are also written in one chunk now, instead of writing them more memory efficient per window. Implementing two different ways of writing the files depending on the format would result in relatively complex code...

Example:

[cholmes]

Looks great! Works great! And man parquet (+brotli) output is so cool. It's like 9x smaller than gpkg / 7x smaller than fgb / 11x smaller than json. It was feeling weird that our vector output was 18 times larger than the raster inference output. With Fiboa/parquet it's pretty much the same (in my single test case with defaults + close interiors, I'm sure it varies)

I do think it could be good to give some nudge to people to use parquet/fiboa output. We could make the output file name not required and just use the input file name and change to parquet.

I was also wondering about fiboa json output - was there a reason you didn't do it?

We could put a warning or something when people pick non-fiboa outputs, to tell them that fiboa output is better for field boundaries.

Anyways, nothing more needed for this PR, it's a great step forward.

[m-mohr]

I was also wondering about fiboa json output - was there a reason you didn't do it?

Honestly, I didn't think about it at 2am ;-)

Looking at it right now, I don't have an exported ready to use method in the fiboa CLI to export to JSON yet. Also, not sure about the replication of the determination fields before we add collection level metadata in fiboa 0.3. The JSON from GDAL should be mostly fiboa compliant anyway apart from the missing fiboa_* fields. Something we can add later though. Until then fiboa create-geojson is your friend :-)

I do think it could be good to give some nudge to people to use parquet/fiboa output. We could make the output file name not required and just use the input file name and change to parquet. [...]
We could put a warning or something when people pick non-fiboa outputs, to tell them that fiboa output is better for field boundaries.

Added :-)