feat: generalize data download to any CoinGecko-listed token

[tiago-hansen]

Context

The price-data pipeline had two disjoint entry points: download_data.py (Binance/Coinbase only) and a one-off prepare_bold_usdc_data.py (CoinGecko). The reCLAMM onboarding skill told users to copy that CoinGecko script every time a non-Binance token appeared. Every downstream consumer only cares about the canonical parquet schema, not the source — so the duplication was avoidable.

What changed

• New quantammsim/utils/data_processing/coingecko_data.py — the CoinGecko source (registry + resolve_cg_id, daily → complete 1-minute canonical grid with the /1440 volume spread) and rebuild_stable_peg. Single source of truth for the COINGECKO_IDS / STABLECOINS / NATIVE_LST registries (moved out of fetch_token_mcaps.py, which now imports them back).
• update_historic_data gains source / cg_id / cg_days. source="auto" (default) runs the exchange waterfall and falls back to CoinGecko only if nothing is found; source="coingecko" fetches directly. The finalization block is extracted into a shared _finalize_historic_data helper so every source produces identical parquet + daily/hourly CSV + plot outputs (CoinGecko tokens now get the daily CSV the old BOLD script never wrote).
• download_data.py — new --source, --cg-id, --cg-days, --peg <STABLE> flags. The BOLD/USDC flow becomes one command:

  python scripts/download_data.py BOLD --cg-id liquity-bold-2 --peg USDC
  

• Removed prepare_bold_usdc_data.py; rewrote reCLAMM SKILL.md Step 1.

Verification

• Grid invariant: CoinGecko frame is an exact 1-min grid, passing the pipeline's == 60000ms assertion.
• Volume invariant: daily resample-sum recovers real daily USD volume (the /1440 spread).
• End-to-end source="coingecko" run wrote parquet (canonical columns, midnight-append), daily+hourly CSVs, and both plots with no KeyError from the new CoinGecko source color.
• Peg builder round-trips at flat $1.00 over the token window with the date/close/Volume USD columns market_features reads.
• fetch_token_mcaps.py, CLI --help, and the single-ticker guard for --cg-id/--peg all pass.

The only path not exercisable offline is the live exchange → CoinGecko auto-fallback (needs a real Binance miss); that branch is a three-line guard feeding the same get_coingecko_data the tests cover.

🤖 Generated with Claude Code