Improve connectors/globs for large datasets

[begelundmuller]

• Add support for ingesting only a subset/sample of data through an extract config option
  • See the proposal for details
• Iteratively ingest files into DuckDB instead of downloading all before ingestion
  • Consider refactoring abstractions around connectors (see code review comment here and/or iterator idea here)
• Add a timeout: [seconds integer] config key for sources
  • Source ingestion should be aborted after timeout, do any relevant cleanup, then return an error
• Add a hive_partitioning: true config option to parse Hive partitions into columns
  • This applies also to regular (non-glob) connectors
  • This can be achieved for DuckDB just by passing HIVE_PARTITIONING=1 to read_parquet
• Add docs for new config options
• Do thorough QA to ensure partial Parquet downloads never lead to degraded performance (versus full download)
• Address code review(s)
• With product input, decide on default config and confirm YAML format for partial ingest
• (Pending) Add support for a latest strategy that lists all files and sorts by last updated timestamp
• (Pending) Add head files of tail folder support