[P1] Add attack effectiveness tiers

[Petsku01]

Problem

All attacks treated equally, but some are noise.

Tasks

• Tier attacks by historical success rate
• Add effectiveness_tier to attack metadata (high/medium/low/untested)
• Allow filtering by tier in CLI

Acceptance Criteria

• Users can focus on high-signal attacks

[Petsku01]

Triage 2026-09-14, verified against main (79aafd0): confirmed open, no tier data or filtering in code.

• effectiveness_tier — 0 hits anywhere
• psg/catalog_validator.py already accepts optional tier on each attack (OPTIONAL_FIELDS includes it), so the schema is ready — what's missing is (a) the tier data itself and (b) CLI filtering
• Filtering: no tier references in psg/cli.py or psg/attack_sets.py

Suggested build order: compute historical success rates from existing results/ runs + the reframing study (both have per-attack outcomes) → write effectiveness_tier into catalog metadata (high/medium/low/untested) → add --tier high[,medium] filter to psg scan and psg benchmark.