v3.0.0-rc.9

Breaking Changes

• Cached::Error and ConcurrentCacheBase::Error are now bounded by
  std::error::Error + Send + Sync + 'static. Custom store implementations whose error
  type does not implement std::error::Error (or is not Send/Sync/'static) must update
  their type Error. All built-in stores (Infallible, CacheSetError, RedisCacheError,
  RedbCacheError) already satisfy the bound. See the migration guide for detection and remediation.
• ConcurrentCached::cache_contains and ConcurrentCachedAsync::async_cache_contains are now
  required methods, no longer defaulted, and no longer carry a V: Clone (async: V: Clone + Send)
  bound. External implementors of ConcurrentCached / ConcurrentCachedAsync must add these
  methods; the previous defaulted version only ever shipped in 3.0.0 rcs. Rationale: the V: Clone
  bound blocked contains for non-Clone value types and could not be relaxed after 3.0.0 without
  another breaking change. ConcurrentCachedExt::contains also drops its V: Clone bound.
• TtlSortedCache::set_ttl inherent method removed. Runtime TTL controls are now only on the
  traits: CacheTtl for single-owner stores, ConcurrentCacheTtl for concurrent stores. Migration:
  use cached::CacheTtl; and the same cache.set_ttl(d) call compiles with identical semantics.
• CachedExt::contains now delegates to Cached::cache_contains (new, defaulted; built-ins
  override with a peek-based implementation) instead of cache_get. As a result contains no
  longer counts a hit/miss, promotes LRU recency, or refreshes TTL on refresh-on-hit stores, and
  reports expired entries as absent. This only affects the rc surface; the trait-level default of
  Cached::cache_contains remains get-based for third-party stores that do not override it.

Added

• CachedPeek::peek: ergonomic alias for cache_peek.
• CloneCached::peek_with_expiry_status: ergonomic alias for cache_peek_with_expiry_status.
• ConcurrentCloneCached::peek_with_expiry_status: ergonomic alias for
  cache_peek_with_expiry_status on the concurrent clone trait.
• CachedExt::reset: ergonomic alias for Cached::cache_reset, mirroring the existing
  ConcurrentCachedExt::reset.
• Cached::cache_contains<Q>(&mut self, k: &Q) -> bool: defaulted presence check on the core
  single-owner trait (default get-based; built-in stores override with a peek-based implementation).
  CachedExt::contains delegates to it, giving contains both spellings on both trait families.
• ConcurrentCached::cache_contains: required presence check returning Result<bool, Self::Error>,
  no V: Clone bound. The built-in sharded stores use a peek-based implementation (read lock, no
  clone, no recency update, no hit/miss metrics); RedisCache / RedbCache
  use a get-based implementation. External implementors of ConcurrentCached must add this method.
• ConcurrentCachedAsync::async_cache_contains: required async counterpart of cache_contains,
  no V: Clone + Send bound; same impl split as the sync method. External implementors of
  ConcurrentCachedAsync must add this method.
• ConcurrentCachedExt::contains: ergonomic alias for cache_contains, no V: Clone bound.
• Inherent contains(&self, &K) -> bool on all six sharded stores: peek-based, infallible,
  takes call-site priority over ConcurrentCachedExt::contains (consistent with the other
  inherent shims like get/set/reset).
• ExpiringLruCache::iter_order / key_order / value_order: recency-order introspection
  matching LruCache (plain (K, V) / K / V shapes; expired entries excluded), completing
  the LRU-family parity started with retain.
• #[must_use] on ConcurrentCached::cache_remove and cache_remove_entry, their
  ConcurrentCachedAsync counterparts async_cache_remove / async_cache_remove_entry, and the
  ConcurrentCachedExt aliases remove / remove_entry.
• #[must_use] on RedbCacheError::is_deserialization, matching RedisCacheError::is_deserialization.
• #[cfg_attr(docsrs, doc(cfg(feature = "time_stores")))] on the ShardedTtlCache* and
  ShardedLruTtlCache* re-exports in src/stores/sharded/mod.rs.

Changed

• encode_ttl/decode_ttl deduplicated from src/stores/sharded/ttl.rs and
  src/stores/sharded/lru_ttl.rs into a single pub(crate) copy in
  src/stores/sharded/mod.rs; both files now import from there. No behavior change.

Documentation

• Prelude doc corrected: it re-exports both traits and the CacheMetrics struct, not traits only.
• The compile_fail snippet for cache_get_or_set_with includes a leading comment making its
  non-compiling intent clear in the generated README (where compile_fail renders as plain Rust).
• docs/migrations/2.0-to-3.0-human.md: added section on Return<T> field privatization
  (r.value / r.was_cached -> deref / into_inner() / was_cached()).
• docs/migrations/2.0-to-3.0.md and 2.0-to-3.0-human.md: added the Cached::Error and
  ConcurrentCacheBase::Error bound as a breaking change entry.
• specs/traits-core.md and specs/traits-concurrent.md: updated to record Error bounds,
  peek aliases, reset, cache_contains/async_cache_contains/contains.
• docs/migrations/2.0-to-3.0.md and 2.0-to-3.0-human.md: added entries for required
  cache_contains/async_cache_contains, TtlSortedCache::set_ttl inherent removal, and the
  single-owner contains semantic change.
• specs/traits-core.md: added Cached::cache_contains defaulted method and updated
  CachedExt::contains delegation.
• specs/traits-concurrent.md: updated cache_contains/async_cache_contains to required,
  unbounded; updated ext alias and IO store impls.
• specs/store-sharded.md: added inherent contains to the sharded store inherent-shim list.