Replace osm2orc with osm-pbf-parquet

[mojodna]

https://github.com/brad-richardson/osm-pbf-parquet

Pros:

• much faster
• produces Parquet
• fixes (or will fix) schema warts like Decimal → double
• partitioned by type (by default)
• not sorted (will help with skew

Cons:

• produces Parquet (not backward-compatible, but that's an opportunity to be taken)
• not sorted
• produces multiple files (but if we don't assume that people want to pass around single files, that should be fine; if that's needed, there may be tricks where we can concatenate file parts as-is without needing to rewrite the whole thing)
• no support for changesets (yet)

Unknowns:

• memory footprint (estimate needed to tune Fargate task definitions appropriately)
• can it stream the PBF from a remote source, or does it assume that it's local? (needed to provision disk appropriately; the current osm2orc tasks don't use any local disk as part of processing)

[ruanqizhen]

Hi Seth, I was wondering if it would be feasible to copy the new result to the us-west-2 region on AWS. This is due to the fact that it is the only region accessible from an Overture account.

[mojodna]

As part of #51, I'd like to switch regions for the default bucket, which would address the problem for Overture. However, that's probably at least a few months out.

In the meantime, it may make sense to create an Overture AWS account where the region restriction doesn't apply and where we can register a job that copies ORC files into an internal Overture bucket. Let's chat about this on Slack.