refactor: replace filteredServerCache with generic syncutil.TTLCache

[Copilot]

internal/server/routed.go contained a ~70-line hand-rolled filteredServerCache that duplicated the get-or-create pattern already present in internal/syncutil/cache.go, while adding TTL eviction and LRU-cap that the generic helper lacked.

Changes

• internal/syncutil/ttl_cache.go — New generic TTLCache[K comparable, V any] with lazy TTL eviction and LRU-cap eviction on every GetOrCreate call. Includes an injectable clock (newTTLCacheWithClock) for deterministic testing.

• internal/server/routed.go — Drops filteredServerCache, filteredServerEntry, newFilteredServerCache, and the getOrCreate method. Replaced with syncutil.NewTTLCache[string, *sdk.Server]:

  // Before: ~70-line hand-rolled struct
  serverCache := newFilteredServerCache(sessionTimeout)
  return serverCache.getOrCreate(backendID, sessionID, func() *sdk.Server { ... })
  
  // After: generic TTLCache
  serverCache := syncutil.NewTTLCache[string, *sdk.Server](sessionTimeout, filteredServerCacheMaxSize)
  cacheKey := fmt.Sprintf("%s/%s", backendID, sessionID)
  return serverCache.GetOrCreate(cacheKey, func() *sdk.Server { ... })

• internal/server/session_util.go — truncateCacheKeyForLog moved here alongside truncateSessionID, its natural home.

• internal/syncutil/ttl_cache_test.go — 9 tests covering cache hit/miss, TTL eviction (fake clock + real-time), LRU ordering, last-used refresh on hit, concurrent access, and maxSize=1 edge case.

• internal/server/routed_test.go — TestFilteredServerCache_MaxSize and TestFilteredServerCache_TTLEviction removed; equivalent coverage migrated to syncutil/ttl_cache_test.go.