perf: fix index node cache thrashing on native + entity attr caching

[tiensonqin]

Summary

Fixes the native daemon being ~2× slower than the Melange/JS daemon on read endpoints. Profiling a warm daemon showed handler CPU dominated by sqlite reloads + transit decode of storage-backed index nodes.

Root cause: persistent-sorted-set caches restored nodes behind OCaml weak refs, which are cleared on every major GC collection. Under any allocation pressure the cache thrashes, so every index slice re-reads nodes from sqlite and re-transit-decodes datoms. Upstream relies on js/WeakRef, which survives V8 minor GCs — the effective cache lifetime is much longer there.

Changes:

• Platform.strong_index_node_cache (new platform flag): true on native, false on melange/jsoo. Storage.settings_of_root overrides the stored ref-type to Strong on native only — an in-memory cache policy; stored metadata unchanged.
• Entity.lazy_entity caches forward-attr lookups per entity (lookup_cache on the materialized assoc list) instead of re-scanning datoms_by_entity per attribute; forward_entity_attr single-attr scan helper dropped as unused.

Measured on macOS arm64, thread-api/api-list-pages over a 3000-block graph (60 calls, p50): 12.3ms → ~4ms daemon-internal handler ~10ms → ~1.9ms. get-page-data on a 12000-block graph: 284ms vs 715ms in the JS daemon. test_entity and the rest of the suite still pass.

Link to Devin session: https://app.devin.ai/sessions/f641a5e9845f45108c4a98cb82e268d3
Open in Devin Desktop: https://app.devin.ai/desktop/session/f641a5e9845f45108c4a98cb82e268d3?variant=devin
Requested by: @tiensonqin

---

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin resolved all 2 findings on 8be3288

Fixed by Devin (2)

• Native restores permanently change cache policy
• Unrelated reference blocks scalar lookups

View all findings in Devin Review