Revamp strategy: switch live deployment to QQQ hourly trend, fix repl…

[utkarshp845]

…ay and data bugs

Why the live BTC bot made no valuable trades (see docs/strategy_revamp_2026-07.md):

• The 5m filter stack fired 2 trades in 120 days of real bars, both losses
• $45 positions cannot clear Alpaca crypto's ~0.6% round-trip friction
• BTC fell 46% over the replay year and Alpaca cannot short crypto

Changes:

• config/live_spy.env + paper_spy.env: QQQ hourly trend-following, ~90% notional fractional sizing, daily-EMA regime gate, wide trails, multi-day holds. Replay 2023-08..2026-07: +37% on $150, PF 1.76, max DD 6.7%, positive every calendar year, robust to 3x slippage
• config/live_btc.env + paper_btc.env: defensive uptrend-only trend config (0 trades in the 2025-26 bear replay, mildly positive in the 2024-25 bull)
• bot/trade_controls.py: replay now resets the consecutive-loss streak on ET day rollover like the live store; previously every backtest locked out entries permanently after the streak limit, invalidating prior research
• bot/broker_alpaca.py: get_recent_bars lookback now scales with timeframe and session hours; the fixed 7-day window could never warm up hourly/daily regime indicators
• bot/profile.py: profile env files now win over market defaults so configs can set SYMBOL/overnight behavior
• docker-compose, deploy script, GitHub workflow: default market btc -> spy (BTC services remain as trade-btc/paper-btc/research-btc)
• tests: updated profile contract expectations, added regression tests for the loss-streak reset and profile precedence (66 passed)