fix: unify simulated cache handling and expose usage to proxy clients

[zhoudashuaibi]

This PR addresses two related issues we hit while running Aether in production:

1. Unlimited wallet balance was still rejected after the wallet state was updated
   When a key's wallet was initially cached as BalanceDenied and then changed to unlimited balance mode, the gateway kept returning the stale cached rejection. This change makes the refresh logic re-check wallet rejections so that unlimited wallets are allowed through instead of being blocked by a stale BalanceDenied state.

2. Simulated cache was not applied consistently and was invisible to proxy clients
   The existing kiro simulated cache only covered a subset of providers. We added a provider-level simulated_cache configuration that applies to all providers and overrides any native cache reporting when enabled. We also wired the simulated cache hit/miss counts through to proxy clients via response metadata.

What's changed

• apps/aether-gateway/src/control/auth/resolution.rs

  • Re-validate wallet rejections during auth context refresh so unlimited wallets can recover from a previous BalanceDenied state.
  • Add test coverage for the unlimited-wallet recovery path.

• apps/aether-gateway/src/execution_runtime/kiro_cache.rs

  • Add SimulatedCacheConfig and provider-level simulated_cache_enabled / simulated_cache_min_percent / simulated_cache_max_percent context fields.

• apps/aether-gateway/src/execution_runtime/simulated_cache.rs (new)

  • Centralized simulated cache implementation that forces cache overrides for providers regardless of native cache support.

• apps/aether-gateway/src/execution_runtime/stream/execution.rs & sync/execution.rs

  • Use simulated cache uniformly and expose cache usage back to proxy clients.

• crates/aether-usage-runtime/src/request_metadata.rs & write.rs

  • Carry simulated cache usage through usage tracking so clients can see it.

• Admin provider write/normalize paths and frontend provider form

  • Allow configuring simulated cache parameters on each provider.

• Usage repositories (memory / postgres / sqlite)

  • Store simulated cache read/write tokens and add test coverage.

Notes for reviewers

• This PR intentionally combines the unlimited-wallet fix and the simulated-cache unification because the second commit builds on the usage-tracking changes introduced in the first one.
• We have been running this internally for a few days and it resolves both the "insufficient balance" false positives for unlimited wallets and the inconsistent cache reporting across providers.