Issue #1553: perf: avoid loading all facts for active-task projection/injection

Status: WIP scaffold PR only — implementation pending.

Source Issue

Issue: #1553
URL: https://github.com/markus-lassfolk/openclaw-hybrid-memory/issues/1553
State: OPEN (snapshot: 2026-05-21)
Priority label: priority:high
Labels: enhancement, priority:high, issue/stage/enriching

Acceptance Criteria / Issue Body

Summary

Active-task projection/injection appears to load all active facts and then filter to project/task facts in memory. With a large facts database, this causes unnecessary SQLite work, JS allocation churn, and memory pressure during every prompt build.

Evidence / suspected code path

The active-task path appears to do a broad fetch before filtering:

const facts = factsDb
  .getAll({ scopeFilter })
  .filter((fact) => fact.category === TASK_LEDGER_CATEGORY)
  .slice(0, factLimit);

On Maeve, the gateway logs reported:

memory-hybrid: injecting 100 active task(s) from category:project facts (87 stale)

The facts.db file is ~421 MB and active fact volume is high. Reading all facts on prompt build is therefore expensive even if only a small subset is needed.

Impact

Prompt build performs unnecessary DB reads and allocations.
Stale/duplicate active-task rows amplify the cost.
In high-concurrency periods, this can add memory churn and latency.
It likely interacts badly with the separate duplicate SQLite-handle/native-RSS issue.

Acceptance criteria

Active-task selection should query only category='project' / task-ledger rows at the SQL level.
Query should select only relevant columns/keys for task reconstruction.
Query should apply candidate limits and terminal-status/staleness filters as early as possible.
Prompt injection should not require hydrating the entire facts table.
Add timing/row-count logs around active-task selection:
- SQL rows scanned/fetched
- active candidates
- stale skipped
- duplicates collapsed
- injected count
- elapsed ms

Suggested fix direction

Add FactsDB.getProjectFacts() or getFactsByCategory(category, options) with indexes.
Ensure category/key/entity indexes exist and are used.
Build active-task rows from latest facts per entity/key rather than all fact history.
Consider a materialized active-task table/view updated by active_task_checkpoint.

Stewardship Notes

This placeholder document exists to create a draft PR branch for Issue Stewardship tracking. The implementation work remains pending and should replace or extend this scaffold with the actual fix and verification evidence.