turborepo

56329a6d - perf: Use `Arc<str>` for task dependency hashes to avoid heap clones (#11962)

perf: Use `Arc<str>` for task dependency hashes to avoid heap clones (#11962) ## Summary - Store task hashes as `Arc<str>` in `TaskHashTrackerState` instead of `String`, so that `calculate_dependency_hashes` clones a ref-counted pointer (atomic increment) instead of heap-allocating a new `String` for each dependency hash lookup. ## Why `queue_task` is called once per task in the topological dispatch loop. For each task, `calculate_dependency_hashes` looks up every dependency's hash and clones it into a `Vec`. On large monorepos like `api` (1687 tasks, ~3 deps each), that's ~5000 `String` heap allocations per run — all for 16-char hex strings that are already stored in the tracker and never mutated. Switching to `Arc<str>` makes each "clone" a pointer-width copy + atomic ref count increment instead of a heap allocation + memcpy. ## Testing - All existing pinned-hash tests pass (the capnp serialization produces identical bytes since `Arc<str>` derefs to `str`). - Added `task_hashable_multiple_dependency_hashes` test with a pinned hash value to guard against serialization regressions. - Profiled on the our biggest repo (1039 packages, 1687 tasks): `queue_task` self-time dropped from ~208ms to ~193ms.