feat(aggregation): native MySQL/SQLite bucketing + ad-hoc cache (#1609, #1610)

[rubenvdlinde]

Summary

Picks up two of the five deferred follow-ups filed against the time-bucket aggregation primitive (#1611):

• MySQL / SQLite native time-bucket aggregation #1609 — Native MySQL/SQLite time-bucket. AggregationRunner::tryNativeAggregation() now detects MySQL/MariaDB and SQLite alongside PostgreSQL and emits the matching native bucketing expression (DATE_FORMAT / strftime) with the correct identifier quoting per engine (backticks on MySQL, double-quotes on Postgres/SQLite). ISO-Monday week-start carries across all three platforms; quarter buckets use the matching CASE / CONCAT form. The response backend field reports mysql / sqlite / postgres / php-fallback so callers can observe which path served the request.
• Cache ad-hoc time-bucket aggregations in AggregationCache #1610 — Read-through cache for ad-hoc queries. AggregationCache gains getAdhoc() / setAdhoc() wrappers that derive the cache name slot from 'adhoc:'.sha1(json_encode(AggregationQuery::toArray())). AggregationQuery::toArray() ksort-sorts filter sub-arrays so {a, b} and {b, a} hash identically. runAdhoc() wires the cache read-through; the existing AggregationCacheInvalidationListener covers ad-hoc entries unchanged because they share the underlying openregister_aggregations distributed cache.

The remaining three follow-ups (#1606 multi-field groupBy, #1607 cumulative/rolling windows, #1608 multi-metric) require a value-object refactor on AggregationQuery plus knock-on translator changes (Solr / ES / Postgres / PHP fallback / GraphQL types) and are deferred to separate opsx cycles. Design notes are recorded in openspec/changes/add-aggregation-enhancements/design.md so the next pickup has the API decisions documented without re-running the discovery work.

Depends on

• feat(aggregation): ad-hoc time-bucket aggregation across REST + GraphQL #1611 (add-time-bucket-aggregation) — this PR extends the runner that lands there. Until feat(aggregation): ad-hoc time-bucket aggregation across REST + GraphQL #1611 merges, the diff here will include feat(aggregation): ad-hoc time-bucket aggregation across REST + GraphQL #1611's commits; the diff collapses naturally once feat(aggregation): ad-hoc time-bucket aggregation across REST + GraphQL #1611 lands and this branch rebases.

API decisions

• Backend annotation surface. Response backend is now one of postgres / mysql / sqlite / php-fallback (was postgres / php-fallback). Additive — clients that branch on "postgres" keep working; clients that previously assumed "php-fallback" === !postgres learn three new values they can also treat as "native fast path".
• Cache key derivation. Ad-hoc cache key is agg:{registerSlug}:{schemaSlug}:adhoc:{sha1(query.toArray())}:{filterHash}:{rbacHash}. The adhoc: prefix keeps ad-hoc and named entries visually distinct in cache dumps without partitioning the backing store.
• Cache TTL. 60 s, same as the existing named-aggregation cache. Shares the AggregationCache::TTL constant.
• Invalidation surface. No changes to AggregationCacheInvalidationListener. The existing global ICache::clear() on lifecycle events covers ad-hoc entries because they share the cache namespace.
• Filter stability. AggregationQuery::toArray() ksort-sorts both the top-level filter map and every operator sub-array ({gt, lte} and {lte, gt} hash identically). Cache hit rate stays high under client filter-key reordering.

What ships

• 5 of 5 task sections done (see openspec/changes/add-aggregation-enhancements/tasks.md).
• 9 new tests across the 2 new test files; 7 new tests added to existing AggregationQueryTest.php + AggregationCacheTest.php. All 113 aggregation unit tests pass.
• Quality gates clean on touched lib files: PHPCS strict / PHPMD strict / Psalm strict / PHPStan level 8.
• Docs updated in docs/technical/aggregation-api.md (backend matrix + cache section) and openspec/platform-capabilities.md (aggregations row).

What stays deferred

• Multi-field groupBy on ad-hoc aggregation #1606 multi-field groupBy
• Running / cumulative aggregates on time-bucket primitive #1607 cumulative / rolling windows
• Multi-metric ad-hoc aggregation (count + sum in one request) #1608 multi-metric per request

Each gets a design-only note in design.md. No new GitHub issues to file — the existing #1606..#1608 are the long-running trackers.

Test plan

• CI green on the new branch (PHPCS / PHPMD / PHPStan / Psalm / PHPUnit).
• After feat(aggregation): ad-hoc time-bucket aggregation across REST + GraphQL #1611 lands and this branch rebases, run the /timeseries endpoint twice against the dev container's MySQL service profile, confirm backend: "mysql" and cached: true on the second call.
• Insert a row into the test schema, call once more, confirm cached: false (eviction listener fired).
• Repeat against SQLite via env override on the same dev container.