fix: decouple warmup_minutes from HLCV cache hash for better reuse

[enarjord]

Summary

• Configs that differ only in trading parameters (EMA spans, warmup ratio) now share the same backtest HLCV cache slot, avoiding redundant re-downloads of candlestick data
• The cache uses a ratchet-up strategy: warmup sufficiency is checked at load time, and the cache is overwritten only when a larger warmup is needed
• Backward-compatible with existing caches (missing cache_meta.json treated as warmup=0)

Problem

When backtesting two configs that share identical backtest.*, live.approved_coins, and exchange settings but differ in bot.long (e.g., different EMA spans), the HLCV data was re-downloaded from scratch. This happened because warmup_minutes—derived from the max EMA span—was included in the cache hash. Different EMA spans → different warmup → different hash → cache miss → full re-download.

Example with two real configs:

• template_candidate.json: entry_volatility_ema_span_hours=672 → warmup=36,288 min (25.2 days)
• template_current.json: entry_volatility_ema_span_hours=2490 → warmup=44,820 min (31.1 days)
• Same coins, same dates, same exchange → but produced different cache hashes

Solution

1. get_cache_hash() — removed warmup_minutes from the hash. The key now depends only on data-defining parameters (coins, dates, exchange, coin age, gap tolerance, sources).

2. load_coins_hlcvs_from_cache() — new warmup_minutes parameter. Reads cache_meta.json and checks cached_warmup >= needed_warmup. Returns None if insufficient.

3. save_coins_hlcvs_to_cache() — new warmup_minutes parameter. Writes cache_meta.json. Removed the early-return guard so that re-fetches with larger warmup can overwrite the same cache slot.

4. prepare_hlcvs_mss() — threads backtest_warmup_minutes through to both load and save.

Ratchet behavior

Run	Config warmup	Cached warmup	Result
1	36,288	—	Miss → fetch → save (cached=36,288)
2	36,288	36,288	Hit
3	44,820	36,288	Miss (insufficient) → fetch → save (cached=44,820)
4	36,288	44,820	Hit (44,820 >= 36,288)
5	44,820	44,820	Hit

Test plan

• 16 new tests in tests/test_cache_warmup_reuse.py covering:
  • Cache hash independence from EMA spans and warmup ratio
  • Non-warmup fields (coins, dates) still differentiate hashes
  • Warmup sufficiency gate (hit/miss/boundary)
  • Legacy cache compatibility (no cache_meta.json)
  • Corrupt cache_meta.json fallback
  • Save persists cache_meta.json (uncompressed + compressed)
  • Overwrite behavior (ratchet-up)
  • Full end-to-end ratchet scenario
• Full test suite: 744 passed, 118 skipped, 0 failures

🤖 Generated with Claude Code

[enarjord]

Superseded by #564, which contains only the targeted HLCV cache warmup reuse changes on top of current master.