perf: add a cached Trino benchmark variant

[bill-ph]

The frozen perf suite currently measures Trino with filesystem caching disabled. Add a separately labeled trino_cached target using the same registered DuckLake tables, SQL, warm-up/measurement counts, and 3-CPU/12Gi worker budget.

Scheduled runs and manual posthog_frozen_perf runs execute baseline and cached variants serially in separate disposable namespaces. The cached variant enables Alluxio filesystem caching with a 16GB cache per node and preserves entries between iterations. A default-off provisioner flag controls newly created catalogs. Existing image overrides are rejected for these comparisons so older control planes cannot silently ignore that flag. The original baseline targets remain unchanged.

This is filesystem caching, including cached data/footer/index bytes; the pinned DuckLake connector does not use the separate heap metadata-cache tier, and no query-result cache is enabled. Cache budgets, eviction, cleanup, and standalone cached-run instructions are documented.

Validation:

• just test-perf, just test-scenario, just test-controlplane-k8s, and the tests/mw-dev harness tests pass.
• Exact pinned Trino image smoke test over a synthetic registered DuckLake table: repeated query returned the same result, with 2,663,270 bytes served from cache and zero additional external reads according to cache-manager JMX counters.
• bash -n and git diff --check pass; review finding about image overrides addressed.
• just lint reports six existing SA4023 diagnostics, reproduced unchanged on clean origin/main with the same toolchain.

The smoke test validates single-node local-file caching. The full distributed S3 perf comparison has not yet been run.

[github-actions]

Test Impact Plan

Deterministic summary of how this PR changes tests, CI runners, and coverage-risk signals.

Summary

Area	Added	Changed	Deleted
Test files	4	23	0
E2E/journey files	0	0	0
Workflow files	0	1	0

Signals

• Test cases: +10 / -0
• Assertions: +53 / -9
• Skips or known failures added: 0
• Workflow continue-on-error added: 0
• Workflow path filters added: 0
• Test commands removed from justfile: 0
• E2E/journey retry lines added: 0

Coverage risk: neutral or increased

No coverage-reduction warnings detected.