chore: avoid hammering store at ntx builder startup

[SantiagoPittella]

closes #1952

This PR:

• Trusts local committed account state on restart and only fetches the unconsumed_notes delta per known account, dropping the per-account GetNetworkAccountDetailsById call.
• Adds a configurable hydration semaphore in StoreClient (defaulted to 4) that bounds concurrent in-flight startup RPCs.
• Adds a separate store_sync_checkpoint column on chain_state (required a new SQL migration), distinct from the mempool-driven block_num.

I tried it by:

1. initializing the main version of the node.
2. Started multiple network monitors to create network accounts.
3. Stoped everything.
4. Installed and initialized the new version with the existing data (i.e. I didn't re-run bootstrap), which ran the migration
5. Everything continue working as expected.

[sergerad]

Do we know how slow things could be because of this?

[SantiagoPittella]

Not really, because my local DB isn't big enough to actually measure this. It would be really nice to have a copy of the DB in a server with similar specs to the devnet/testnet ones to compare it. Or at least running it locally with the testnet database.

Do you know if I can get access to it? maybe @Mirko-von-Leipzig ?