refactor(datalake): redesign clean result layout, move to columnar crawl format, and support many crawlers per bucket

[thecaffiend]

Background

The result-clean bucket layout and table-naming scheme were built quickly during the bactopia-only era, when there was a single result source. The crawler config prefix becomes the Glue table-name prefix and one crawler walks the entire bucket, so every top-level folder becomes a result_<folder> table sharing one crawl scope, schedule, classifiers, and excludes.

Adding a second result source (the caerbannog consumer ETL in #363) makes this coupling a problem: distinct datasets are forced under one result_ namespace and one crawler with no isolation. The current approach works for the demo but will not scale as result sources multiply.

Scope

• Redesign the clean result bucket layout and table-naming so multiple independent result sources coexist without a shared flat result_ namespace (per-source prefixes/namespacing).
• Move crawled output to a columnar format (Parquet, and evaluate Iceberg) instead of CSV. This is intended as a single coordinated switch across the whole datalake, not a per-source change. (No pre-existing tracking issue was found for the columnar move; folding it in here.)
• Support many crawlers per crawled bucket. Today configure_bucket builds exactly one DataCrawler per bucket from a single crawler: config block; extend the config schema, configure_bucket, and DataCrawler to allow a list of crawlers per bucket so each result source can have its own schedule, classifiers, excludes, and table prefix.

Notes

• Surfaced while scaffolding feat(seqauto): add new raw result etl and update report to show data from it #363 (new seqauto raw-result ETL for the consumer output). That branch deliberately reuses the existing single result-clean crawler with distinct top-level output folders as a low-churn demo path, deferring this redesign here.
• Related: fix how we do custom classifier configuration #8 (custom classifier configuration).

Refs #363