`get_existing_candles` times out on larger PostgreSQL databases because of N+1 queries

[l-j-g]

The get_existing_candles() endpoint can exceed the MCP client's 10-second request timeout when the Jesse candle database contains multiple exchange/symbol datasets.

Observed error:

Network error during existing candles retrieval:
HTTPConnectionPool(host='127.0.0.1', port=9000):
Read timed out. (read timeout=10)

[axlnur]

Reproduced on a production database: PostgreSQL 14.23, candle = 663,146,368 rows (~176 GB), 536 exchange/symbol pairs. The current implementation takes 15-20 minutes, so the MCP tool's 10-second timeout always fires.

A single GROUP BY aggregate (#616) removes the N+1 pattern but still reads all rows in one pass — 3 min 11.6 s on this database, which still misses the 10 s budget. Since PostgreSQL has no native index skip scan before v18, #621 rewrites the query as a recursive CTE that walks the (exchange, symbol, timeframe, timestamp) unique index group by group — O(groups x log N) index probes instead of O(N): 25-40 ms warm / ~1.0 s cold on the same data, byte-identical result set.