fix(system-health): detect modern sql.js memory DB paths

[IRakow]

Problem

system_health reports memory as degraded with "Memory store not found — run memory init" on every healthy sql.js install.

Reproduction against v3.5.80:

$ ruflo memory init
$ ruflo memory stats
  initialized: true, backend: sql.js + HNSW, ...
$ (call mcp__ruflo__system_health)
  memory: degraded — "Memory store not found — run memory init"

Cause

v3/@claude-flow/cli/src/mcp-tools/system-tools.ts at line 312 checks only the legacy JSON-backend path:

const memoryDbPath = join(projectCwd, '.claude-flow', 'memory', 'store.json');

The modern sql.js backend — the default on current installs, per @claude-flow/cli/.../memory-initializer.js:366 — persists to .swarm/memory.db. The legacy JSON path never exists on these installs, so the check always reports degraded.

Same file already looks at the sql.js paths elsewhere (statusline-generator.js lines 231-233, 501-503) using exactly .swarm/memory.db, .claude-flow/memory.db, data/memory.db — aligning system_health with that convention.

Fix

Check the modern sql.js paths first, fall back to the legacy JSON path for backwards compatibility. Five-line change, no behavior change when either path exists — only fixes the false-positive degraded status.

Verified

Applied same patch to the built system-tools.js locally on v3.5.80; system_health flips memory from degraded → healthy when .swarm/memory.db exists (the expected state after memory init), and continues to report degraded when no store exists at all.

Notes

• Every Ruflo user on a sql.js install sees this — it's not install-specific.
• Non-functional: memory_stats, memory_store, memory_search all work correctly; only the health reporting was wrong.