feat(solana-indexer): PR 14 PostgresStore persist decoded events + slot watermark

[squadgazzz]

Description

Implements the PostgresStore persistence for the indexer's live path: decoded settlement events and the slot watermark land in the solana.* tables in one transaction. This retires the todo!() stubs on Persistence for the ingest path and gives the decoder a real store to write through in a follow-up.

Runs ahead of the schema being frozen. The solana.* DDL lives in a test-only fixture applied to docker Postgres in the #[ignore] test, not a committed flyway migration, so staging and prod stay clean until the shape settles. When it does, it promotes to database/sql-solana/ plus a compose service, mirroring the pool-indexer.

Changes

• Persistence holds a PgPool and implements persist_events (events plus watermark in one transaction), write_watermark (monotonic), and read_watermark, all runtime sqlx queries.
• Settlement mapping: OrderCreated writes solana.order_pda (off-chain path, solana.orders is the orderbook's row), SettlementFinalized writes solana.settlements plus per-order solana.trades plus the cumulative solana.order_pda counters, OrderClosed/OrderCancelled set the cancellation timestamp. Buffer and admin events are skipped (no MVP table).
• Test-only schema.sql fixture and an #[ignore] docker-Postgres test.
• The finalization and recovery Persistence methods stay todo!() (their own PRs, recovery is out of scope for v0.1).

How to test

docker compose up -d, then cargo nextest run -p solana-indexer --run-ignored all --test-threads 1. Non-ignored tests run without a database.

Notes

• Placeholders mirror PR 7.2: instruction_index and fee_amount on solana.trades, and the buy-side amount, are constants until the decoder resolves them from proposed-solution data.
• The decoder does not call persist_events yet. Wiring it needs a Store trait so the decoder unit tests inject a fake rather than a live database, deferred to a follow-up.
• schema.sql is deliberately not a flyway migration (see Description).