Adds the core faceted search engine, the reviewboard/search/facets/ package,

along with benchmark tooling for performance validation.
Key components:

- DisjointFacetEngine (aggregations.py) builds and executes a single

  Elasticsearch msearch body per request containing: a paginated results

  query, one aggregation sub-query per faceted=True FilterSpec (each

  excluding its own filter clause so bucket counts remain accurate when the

  filter is active — the disjoint property), and one option-count query per

  compound filter choice.

- FilterSpec / FilterBuilder (schema.py, builder.py) provide a

  declarative schema for filter dimensions and translate active filter values

  into Elasticsearch bool.filter clauses. Compound filters (ship-it, issues,

  reviews, file attachments) use custom build_fn callables.

- FacetCache (cache.py) caches aggregation results per group, query, and

  filter state for the lifetime of a request.

- FacetedSearchEngine (engine.py) orchestrates searches across

ReviewRequestSearchGroup, UserSearchGroup, and ReviewGroupSearchGroup.

  For the active group it runs a full search with aggregations; for inactive

  groups it runs a count-only query so sidebar tab totals are accurate.

- ReviewRequestSearchGroup additionally handles the Contents filter, which

  restricts the text query to specific ES fields and optionally runs a parallel

  query against comment indexes to surface review requests with matching

  comments.

- Also adds populate_bench_data management command and bench_manager.py for

  populating and benchmarking persistent datasets at 1k, 10k, 100k, and 1M

  scale across Postgres, MySQL, and MariaDB.

Verified that DisjointFacetEngine builds the correct msearch body structure,

enforces the disjoint property on aggregation sub-queries, and produces exactly

one msearch call per search regardless of how many filters are active. Verified

that FilterBuilder generates correct Elasticsearch clauses for each filter

type and that the permission filter is always injected. Verified that

FacetedSearchEngine routes full searches to the active group and count-only

queries to inactive groups.
Performance benchmarks were run at 10k, 100k, and 1M review requests on

Postgres, MySQL, and MariaDB. A 6-filter faceted search against a 1M-document

index completes in 4.7ms with latency remaining flat across all scales. Full

results are in the attached benchmark report.