feat(v10.5): paper trader — Track 1 simulator with live PancakeSwap quotes

Bridges NRI signals to simulated execution without touching real wallets.
Pure read-only PancakeSwap router quotes, no broadcasts, no private keys.

Architecture:
  - paper_trader.py: 15-min decision cycle reads /api/live, applies
    rotation rules, logs to data/paper_ledger.csv. State in
    data/paper_state.json.
  - Sizing: STRONG_LONG=30%, LONG=15% top + 10% #2, NEUTRAL/AVOID=cash
  - DEFENSIVE_ROTATION trigger (RISK_OFF + top conviction <30) flattens
    everything to cash. Working as designed: current snapshot is RISK_OFF
    so no positions opened.
  - Concentration: 30% single-token cap + first-come-first-served across
    overlapping narrative buckets (e.g. SIREN appears in both Meme and
    AI Agents, gets allocated only once at higher conviction).
  - Filter: ASCII-only symbol guard (skips UTF-8 names with no DEX pair),
    $1M minimum mcap, sort by 7d change descending (rotate-into-strength).

Endpoints (web/app.py):
  - GET /api/paper          state snapshot (bankroll, positions, PnL)
  - GET /api/paper/ledger   recent decisions

Dashboard panel:
  - 4 stat cards (bankroll / cash / positions / realized PnL)
  - Open positions table with live mark prices and PnL
  - Recent decisions ledger (last 8)

Smoke test (TRANSITION regime, mock conviction):
  Meme STRONG_LONG → SIREN $285 (28%)
  AI Tokens LONG    → SKYAI $95
  AI Agents LONG    → AIO $143
  RWA LONG          → BTX $143 + RWA $95
  → 5 BUYs, $762 deployed, $238 cash buffer

Cron nri-paper-trader runs every 15min, logs silently unless trades fire.

Author: asbestos22