fix: protect freshly accepted DIPs rules with durable state

[MoonBoi9001]

TL;DR

Replaces an in-memory, time-based workaround that protected freshly accepted DIPs indexing rules with durable local state, so the protection survives an agent restart and drops the fixed timer. No behaviour change in the normal case — it closes a gap where a freshly paid-for rule could be deleted.

Motivation

When the agent accepts a DIPs indexing agreement on-chain, it creates a local indexing rule for that deployment. A separate cleanup loop deletes DIPs rules that aren't backed by either a pending proposal or an agreement the indexing-payments subgraph reports as active. For a few blocks after acceptance the agreement is neither — it's been accepted, but the subgraph hasn't indexed it yet — so the freshly created rule looks unbacked.

Until now a 300-second in-memory grace window bridged that gap, but it didn't survive an agent restart: a restart during the window could let the cleanup delete a rule the indexer is actively being paid to serve, churning the deployment and its allocation until the subgraph caught up. The agent already records each acceptance durably, so the fix is to keep the rule alive from that record instead.

Summary

• Keep a freshly accepted agreement's rule alive from its durable local accepted record.
• Retire that record once the network data lists its agreement, instead of racing two clocks.
• Fall back to the subgraph's progress as a backstop when an agreement isn't listed yet.
• Skip a single failed retire and retry it next cycle, rather than aborting the cleanup pass.
• Warn and list the stuck deployments when no indexing-payments subgraph is configured.
• Remove the in-memory grace window and its fixed 300-second timer.
• Cover the keep, retire, reap, and missing-subgraph paths with tests.