Fix stats search

[tiborrr]

Why this PR

Opening the dashboard was slow on my instance (~800k docs). Each box that shows search summaries triggered POST /searchStats, and the dashboard fired several of those at once — often the same query more than once. On a database with a large custom_field_value table, a single stats request could take 7–12 seconds even for a default empty query.

Before — slowest query on dashboard load (general profile, fieldCount):

SELECT COUNT(DISTINCT cf.id) AS num
FROM custom_field_value cvf
INNER JOIN custom_field cf ON (cf.id = cvf.field AND cf.coll_id = $1)
WHERE cvf.item_id IN (
  SELECT i.itemid FROM item i
  WHERE i.coll_id = $2 AND i.state IN ($3, $4) AND i.folder_id IS NULL
) AND NOT cf.ftype IN ($5, $6)

~11.7s on ~895k matching items (full table scan of custom_field_value against a huge IN subselect).

Solution

1. Split searchStats into profiles

Introduce statsProfile so callers fetch only what they need:

Profile	count	tagCloud	fieldStats	folderStats	dimension lists	fieldCount / orgCount / …
full	✓	✓	✓	✓	✓	—
general	✓	✓	—	—	—	✓
fields	✓	—	✓	—	—	—

• Backend: POST /sec/item/searchStats/{profile} (path > body statsProfile > default full).
• Dashboard: Fetches general + fields instead of repeated full stats; StatsCache dedupes so boxes sharing a query hit the backend once per profile.
• Backward compatible: POST /searchStats without a profile still returns full.

2. Large-database field-stats optimizations

Three targeted fixes in QItemFieldStats (verified on PostgreSQL with ~895k items):

Issue	Fix
OOM loading ~895k item IDs into the JVM	Count first; only load IDs when match set ≤ 200
fields profile ~12s aggregating text fields the UI never shows	For large match sets, only aggregate money/numeric fields (SearchStatsView filters to sum > 0 anyway)
general.fieldCount ~12s via cvf IN (subselect)	cf-first EXISTS probe using custom_field_value.field index

After: fields aggregation for text-only collectives completes in <1ms; fieldCount uses the fast EXISTS pattern instead of the 11s IN subselect.

API / docs

• New StatsProfile type (full | general | fields).
• OpenAPI: profile matrix, path routes, fieldCount semantics, deprecation note for ?statsProfile= on POST.
• Changelog entry under v0.44.0.

Tests

• SearchStatsTest (H2): each profile omits/includes expected sections; general.fieldCount counts non-numeric field definitions with values.

Manual verification (PostgreSQL)

CI uses H2 only. On a large DB (~900k items):

1. Dashboard loads without OOM.
2. Network tab shows /searchStats/general and /searchStats/fields, not duplicate full stats calls.
3. pg_stat_statements: no multi-second COUNT(DISTINCT cf.id) … cvf.item_id IN (SELECT …) on dashboard load.
4. Numeric custom fields still show sums/avgs when present.

|calls|total_time_seconds|avg_time_ms|query|
|-----|------------------|-----------|-----|
|3|1.415178407|471.72613566666666|SELECT COUNT(i.itemid)  AS num  FROM item i  WHERE (i.coll_id = $1  AND i.state IN ($2 , $3 ) AND i.folder_id is null )|