docs: cover reliable provider workflows

[mattsse]

Summary

• compare live subscriptions, JSON-RPC filter polling, and resumable historical streams
• document canonical block and log streams, reorg handling, pubsub reconnection, and backfill responsibilities
• separate connection retries, request backoff, endpoint fallback, call timeouts, and transaction confirmation timeouts
• map optional RPC namespaces to provider extension traits and feature flags, including direct alloy-provider extensions

Why

Provider pages currently explain how to construct HTTP, WebSocket, and IPC clients but stop before production behavior. They do not tell readers how to resume an indexer, handle reorgs, configure retry boundaries, or discover non-eth RPC methods.

Impact

Applications can choose an event-delivery model explicitly, implement gap and reorg recovery, and configure retries without conflating different failure boundaries. The namespace map gives users and agents the exact feature and trait vocabulary needed to find debug, trace, txpool, engine, admin, Anvil, MEV, and other APIs.