feat: ingest and store the epoch schedule for warmup-aware epoch math

[Mctursh]

Makes the epoch math use the cluster's real schedule instead of the hardcoded mainnet constant.

What it does

• Ingestor: when RPC_URL is set, fetches getEpochSchedule at startup and writes it to a new single-row epoch_schedule table. Best-effort, a failure logs and doesn't block ingestion.
• RPC: reads it once at startup into AppState. The read is outside create_tables(), so a missing table doesn't break existing deployments. It falls back to the mainnet schedule (without_warmup) and logs which path it took.
• getInflationReward derives its default epoch and slot bounds from the stored schedule. getEpochInfo (feat: implement getEpochInfo RPC method #30) and getEpochSchedule (feat: implement getEpochSchedule RPC method #36) are still open, so those are a small follow-up once they merge.

Your two calls on #47

• No new crate: solana-epoch-schedule is already in the tree via solana-sdk, so this just promotes it to a direct dep. Zero added build cost.
• Config for other clusters: went with RPC_URL instead of the config flags you floated on Ingest and store the epoch schedule so the epoch methods are correct off mainnet #47, since the cluster already knows its own schedule. The ingestor just pulls it from whatever cluster RPC_URL points at (devnet/testnet/etc). Can add explicit override flags on top if you'd rather have that knob.

Singleton table
First single-row table in the repo, so worth a look at the shape. ReplacingMergeTree keyed on slots_per_epoch, so re-ingesting the same schedule collapses to one row; cluster and replicated wrap it in Distributed like the rest. One difference from the other tables: they all pass a version column (ReplacingMergeTree(slot)), this one doesn't, which is fine while the schedule is immutable. If you'd want a re-ingest to overwrite instead, that's a version column plus a constant key.

Testing
Unit tests cover the mainnet-matches-old-math and warmup-differs cases. E2E on a local ClickHouse: ingest wrote the row, RPC loaded it, and dropping the table fell back to mainnet and still started.

Closes #47.