fix: add DISABLE_SIGNIFICANCE_FILTER env var to bypass memory filtering

[RyanNg1403]

Summary

• Add DISABLE_SIGNIFICANCE_FILTER=true env var to bypass the isSignificantKnowledge() and isWorkspaceSignificantContent() filters
• Default is false — no behavior change for existing users
• When enabled, all content passes through to embedding/storage instead of being silently rejected by regex heuristics

Problem

The significance filter uses skipPatterns that match common technical terms:

/\b(personal|profile|identity|username|login|password|email|address|phone)\b/i

Content mentioning "email" (email services, SMTP), "address" (IP addresses, memory addresses), "login" (auth systems), etc. is silently discarded. The tool returns extracted: 0, skipped: 1 with no user-facing indication that the filter was the cause — identical to a dedup skip.

Root cause analysis by @MafraAiDev in #263.

Usage

DISABLE_SIGNIFICANCE_FILTER=true

What this does NOT guarantee

Issue #263 has multiple reporters with different setups and potentially different root causes:

• @MafraAiDev confirmed the significance filter as root cause (Qdrant + OpenAI) — this fix directly addresses that
• @temporary-wanderer found a separate OpenAI SDK base64 encoding issue with LM Studio — this fix does not address that
• @Sheld0n-Cooper (original reporter, Gemini API) may have had a Gemini initialization issue — this fix may or may not help

Test plan

• All 40 memory tool tests pass
• pnpm run format:check passes
• Env var follows identical pattern to existing DISABLE_DEFAULT_MEMORY, DISABLE_EMBEDDINGS, etc.
• Default false — zero behavior change unless explicitly opted in

Fixes #263