Upgrade GitHub Actions tooling:
- Promote CodeQL workflow to Advanced setup with multi-language support
  (actions, go, ruby) and explicit build-mode per language
- Replace manual golangci-lint curl install with golangci/golangci-lint-action
  in the lint workflow
- Remove the redundant pre-commit CI workflow
- Bump golangci-lint to v2.12.1 and buf to v1.69.0 across project settings,
  Makefile, and pre-commit hook

Introduce internal/constants/errors.go to centralize HTTP JSON error labels
and messages (ErrorLabel, ErrMegMissingCacheKey,
ErrMsgUnsupportedDistributedBackend), replacing inline string literals
scattered across management_http.go, pkg/backend/dist_http_server.go,
and related tests.

Apply minor whitespace formatting improvements across multiple source files
to satisfy linter rules.

Add benchmark suites for ConcurrentMap, HistogramStatsCollector, and
HyperCache concurrency paths as regression yardsticks ahead of planned
performance optimizations (Phases 1–2). Capture an initial
bench-baseline.txt on Apple M4 Pro for benchstat comparison.

Harden CI and tooling:
- Upgrade test-race to -count=10 -shuffle=on -timeout=15m in both
  Makefile and GitHub Actions workflow for stronger flake detection
- Enable golangci-lint on test files (tests: true)
- Add typecheck (go vet), build, and ci aggregate Makefile targets
- Add bench-baseline capture target for ongoing benchstat comparisons
- Remove stale golangci-lint badge link from README

… packages

Remove the global sync.RWMutex from HyperCache and the InMemory backend,
delegating thread-safety to the eviction algorithm's internal lock and
the sharded ConcurrentMap respectively.

Rewrite HistogramStatsCollector with per-stat atomic aggregates (count,
sum, min, max) and a bounded ring buffer (DefaultSampleCapacity=4096) to
fix the sort-under-RLock data race where GetStats sorted the live backing
slice in-place while concurrent writers appended to it.

Store the DistMemory transport in an atomic.Pointer<distTransportSlot> via
loadTransport/storeTransport helpers to prevent concurrent read/write races
across forwarding, replication, and hinted-handoff paths.

Additional fixes:
- NewConfig now returns (*Config[T], error) instead of panicking on an
  empty backendType; all call-sites updated including examples and benchmarks
- DistMemory.Stop is now idempotent and closes tombStopCh and
  rebalanceStopCh goroutines that were previously leaked
- WorkerPool.Enqueue silently drops jobs after Shutdown instead of sending
  on a closed channel, preventing panics during graceful cache teardown
- Suppress Fiber startup banner in management and dist HTTP servers to
  reduce -count=N test noise
- Fix typo ErrMegMissingCacheKey → ErrMsgMissingCacheKey
- Harden distributed tests with StopOnCleanup, polling loops instead of
  fixed sleeps, and increased heartbeat/merkle timeouts for -race runs

BREAKING CHANGE: NewConfig signature changed from *Config[T] to (*Config[T], error).

…ontention

Add `evictionShardCount` field to `HyperCache` and a corresponding
`WithEvictionShardCount` option (default: 32, matching pkg/cache/v2.ShardCount).

When `evictionShardCount > 1`, `initEvictionAlgorithm` wraps the selected
algorithm in `eviction.Sharded`, aligning each key's data shard with its
eviction shard to eliminate the single global eviction-algorithm mutex.
This trades strict global LRU/LFU ordering for significantly improved
parallel-write throughput (~4x on Apple M4 Pro: SetParallel drops from
~750 ns/op to ~200 ns/op). Values ≤ 1 preserve the previous single-instance
behaviour with strict ordering.

Additional changes:
- Replace `errors.New` with `ewrap.New` for ErrInvalidAddress in cluster/node.go
- Add benchmark result snapshots: bench-baseline-v2, bench-step1, bench-step1-unit, bench-step2
- Add lint baseline snapshot: lint-baseline-v2

… parallelize, remove pkg/cache v1

Migrate all test assertions from github.com/longbridgeapp/assert to
github.com/stretchr/testify/require (promoted from indirect to direct
dependency in go.mod). The longbridgeapp/assert package is fully removed.

Add t.Parallel() to every applicable test function across the codebase,
including hypercache_test.go, pool_test.go, pkg/eviction, pkg/stats,
pkg/cache/v2, and the full tests/ and tests/integration/ trees. Tests
that share a cache instance and depend on insertion order are explicitly
annotated with //nolint:paralleltest.

Remove pkg/cache v1 (cmap.go, item.go and their tests) entirely. The
CAWOLFU eviction algorithm now uses a plain map[string]*CAWOLFUNode
guarded by its own mutex instead of the redundant ConcurrentMap sharding.

Additional modernization:
- Fix golangci.yaml: add path-based exclusions for test-specific linters
  (funlen, cyclop, gocognit, dupl, gocritic, goconst, noinlineerr, revive
  function-length/cognitive-complexity/cyclomatic).
- Rename benchmark package declarations from `tests` to `benchmark`.
- Replace http.Get/client.Get with context-aware http.NewRequestWithContext.
- Use t.Cleanup for deferred Stop/Close calls to surface return errors.
- Wrap sentinel errors with %w; introduce errUnexpectedStatus.
- Rename local `cap` constants to `windowCap` to avoid shadowing the builtin.
- Use net.ListenConfig.Listen instead of net.Listen in allocatePort.
- Add nolint directives for gosec, revive, and prealloc where appropriate.

Replace two `//nolint:gosec` suppressed `uint32(...)` casts in `Sharded`
with explicit calls to `converters.ToUint32` from `github.com/hyp3rd/sectools`.

- `NewSharded`: compute shard mask via `converters.ToUint32(shardCount - 1)`
  and propagate any conversion error instead of silencing the lint warning.
- `Evict`: convert `len(s.shards)` via `converters.ToUint32`, returning a
  zero-value on error rather than casting unchecked.
- Add the `sectools/pkg/converters` import to `sharded.go`.
- Register unsharded in `cspell.config.yaml` to keep spell-check clean.