Efficiency improvements

[khoroshevskyi]

Changes:

• Cache get_stats() with a TTL to avoid running uncached COUNT queries on the bed table on every request to hot API paths (stats, neighbours, list, search)
• Eliminated an N+1 query in get_neighbours() by fetching all neighbour metadata in a single batched query (with annotations eager-loaded) instead of one query per neighbour; stale Qdrant points are now skipped rather than raising
• Eliminated an N+1 in BedAgentBedSet.get_ids_list(): it was refetching each bedset by id and lazy-loading its full bedfile membership just to build the list page. Now builds results directly from the paginated query; bed_ids is left unpopulated on list results (use get(identifier) for a single bedset's member ids)
• get_detailed_stats() no longer reuses a Session after its with block has closed (was forcing 3 extra connection checkouts for _stats_comments/_stats_geo_status/_get_geo_stats); all queries now share one session/transaction
• Replaced the bed_files_info() call inside get_detailed_stats() with a targeted 3-column query, avoiding a full-table FileInfo Pydantic construction (with per-row try/except) for every bed record just to extract number_of_regions/mean_region_width/file_size for histogram binning
• BedAgentBedFile.get_ids_list() (backs /bed/list) had no order_by() on its paginated query, so row order across pages was undefined -- rows could be duplicated or skipped between requests. Now orders by Bed.id.
• get_detailed_stats() crashed with a pydantic ValidationError whenever bed_compliance, data_format, genome_alias, species_name, assay, or cell_line had NULL rows: the GROUP BY queries included the NULL group, producing a None dict key, which FileStats's dict[str, int] fields reject. All six queries now filter out NULLs before grouping.
• Added a denormalized bedfile_count column to bedsets, exposed as BedSetMetadata.bedfile_count. Set once at bedset creation time (membership is write-once; add_bedfile/delete_bedfile are unimplemented), so reads never need to touch bedfile_bedset_relation to know a bedset's size. Requires a DB migration -- see scripts/migrations/2026_07_31_add_bedset_bedfile_count.sql

TODO:

• Version of pepdbagent updated in __version__.py file
• Changelog updated